Wayfinder Paper v1
Uploaded by
Lars ErikssonWayfinder Paper v1
Uploaded by
Lars ErikssonWhitepaper V1.
0, March 2024
xecutive Summary
E
3
Key Elements of the Wayfinder Ecosystem 4
Introduction 6
Protocol Overview 9
AI Agent Fundamentals 11
Agent Overview 11
Agent Context and Workflow 11
The Blockchain Environment and its Agents 13
Challenges for AI Agents in Blockchain Contexts 13
Lack of Context 13
Efficiency 13
Shortcoming of Existing Tooling 14
Advantages of Blockchain Environments 14
Advantages of AI Agents On-Chain 14
Key Elements of Wayfinder 17
Architectural Overview 17
Wayfinder’s Technical Approach & Key Tools 18
Consensus & Technical Considerations 18
Purpose-Built LLMs 19
User-Friendly Interface for Non-Coders 19
Comprehensive Integrations for Existing AI Agent Frameworks 20
Shell Operations and “Explicit Intent” 20
Wayfinder, Shells, and Financial Regulation 21
Shell Wayfinding 21
The Wayfinder Graph 23
Transforming the Graph into a Dynamic Toolchain 25
Wayfinding Path Creation and Wayfinding Path Types 25
Seeding the Network’s Initial Wayfinding Paths 26
Crowd-Sourced Establishment of New Wayfinding Paths 26
Trusted Participant Incentives 28
Wayfinding Path Background and Security 28
Establishing Private Wayfinding Paths 30
Wayfinding Bounty Program 30
Overview of Wayfinder Shells 32
Wayfinder Shells and Private Key Management 33
Shells and Wayfinding Path Discovery 33
Wayfinding Path Risks 35
Catastrophic Planning Failures 36
Creating and Managing AI Agents 38
Genesis of a Shell 39
1
Memory and Advanced Features 1
4
Generation and Customization of Memories 41
Enhancing Context with External Data 42
Wayfinder’s Native Token 44
Potential Token Supply and Distribution 46
Primary Token Functions 48
Potential Ways to Earn Wayfinder Native Tokens 49
Wayfinder’s Initial Use Cases 50
Wayfinder’s Substrate for Provably Scarce Generated Resources 50
Application-Agnostic AI Agents 50
NFT Minting Assistant 50
Smart Trading Bot 51
Governance 51
Governance Overview 52
Conclusion 54
2
Executive Summary
ayfinder is a new artificial intelligence (AI) focused omni-chain tool enabling
W
user-owned, autonomous AI agents to securely and efficiently navigate within and
across blockchain ecosystems and applications while independently transacting
assets they control via dedicated Web3 wallets. Wayfinder achieves this functionality
by relying upon novel solutions to a number of widely recognized yet previously
unaddressed technical challenges.
ayfinder’s architecture incorporates a continuously expanding,
W
community-developed network of wayfinding paths, representing unique edges and
nodes of a larger map of blockchain-based destinations. These destinations
collectively form a growing “ecosystem graph” of smart contracts. Wayfinder’s
wayfinding paths and ecosystem graph enable AI agents to securely and efficiently
navigate blockchain environments–in effect creating ‘smart maps’ of smart contracts
that increase the accessibility of smart contracts for AI agents–and to autonomously
transact in a manner that has not previously been possible. Combined with the ability
of AI agents within Wayfinder to remember and learn from their past experiences, as
well as the experiences of other agents within the network, Wayfinder unlocks myriad
possibilities and capabilities for self-improvement and continuous learning within
decentralized ecosystems.
iven the significance of Wayfinder’s innovations, and the numerous ways it promises
G
to enable novel capabilities for AI agents, its potential impacts can be considered
across a number of domains. These include improvements relative to the general
accessibility of blockchain and Web3 ecosystems for both skilled and less technical
users, meaningful contributions to the development of cross-chain interoperability
solutions that substantially reduce the complexity of blockchain environments, as well
as innovations relative to gaming environments and related tooling, to name merely a
few.
In addition to being an important general innovation within blockchain environments
and AI ecosystems, Wayfinder also represents a significant extension of the
decentralizedEchelon Primeecosystem, enabling Parallel Studios’ newColony
AI-based survival as one of the first examples of Wayfinder’s ability to serve as an
underlying set of tools powering games across multiple genres.
he information contained in this paper is intended to serve as an illustrative starting
T
point for discussions of how the Wayfinder ecosystem may evolve. All structures and
decisions described in this whitepaper are ultimately subject to community
stakeholder input and community governance, and may evolve differently as a result of
decisions made by community stakeholders and the larger Wayfinder community.
3
Key Elements of the Wayfinder Ecosystem
● W
ayfinder is a new AI tool optimized for blockchain environments that was
developed as core infrastructure forColony,ParallelStudios’ forthcoming
blockchain-based and AI-powered strategic survival simulation game.
● W
ayfinder’s technological innovations are underpinned by powerful new research
enabling autonomous and continuously learning AI agents to navigate blockchain
ecosystems, securely control dedicated Web3 wallets, and autonomously transact
assets they control.
● W
ayfinder also has the potential to introduce important solutions enabling AI agents
to securely manage private keys for digital wallets, a critical step in allowing
autonomous Web3 transactions.
● W
ayfinder Shells meaningfully contribute to ongoing blockchain interoperability
efforts, substantially reducing the user-facing operational complexity of blockchain
environments.
● W
ayfinder Shells are anticipated to be created by spending Wayfinder’s proposed
native digital asset, pending community approval of a governance proposal to
develop such a token. Once created, shells benefit from the network’s shared
tooling, knowledge library, and the collective intelligence generated by the efforts
of all the network’s shells.
● W
ayfinder’s proposed native token secures shell access to blockchain navigation via
use of the network’s wayfinding paths and smart maps that collectively form an
expansive graph of blockchain ecosystems optimized for AI agents.
● W
ayfinder’s proposed native token could also be used to reward bounties
encouraging crowdsourced wayfinding, to maintain the network and ensure its
security, to compensate wayfinding path creators for path usage, to serve as a
means of payment, and function as the network’s governance vehicle.
● G
iven Wayfinder’s significance as a general purpose AI tool for blockchain
environments, a dedicated token for Wayfinder is anticipated to enable the network
to operate more efficiently and to be fully valued as a reflection of the importance
of the network’s innovations.
● T
he Wayfinder Foundation has proposed, subject to community approval, to entrust
the initial governance of the Wayfinder ecosystem to the Echelon Prime
Foundation’s community, PRIME holders and potentially other ecosystem partners to
establish the initial governance process that will evaluate proposals which create a
dedicated token and long term governance for the Wayfinder community.
● P
roposed Token Supply and Distribution:Upon communityapproval & ratification,
Wayfinder's native token will launch with a maximum authorized supply of 1 billion
tokens. The initial allocation is anticipated to be as follows:
4
*Note that the investor token allocation pool amount is not subject to community governance.
5
Introduction
he world is changing—breakthroughs in AI are bringing rapid increases in
T
productivity, and expectations for AI’s increased impact only continue to grow. Yet
despite widespread claims concerning the transformative impacts that AI will soon
have across the economy, and society as a whole, the reality remains that current uses
for AI remain largely confined to two areas: information search and content creation.
This seemingly limited application mirrors the trajectory of earlier tech innovations that
generated considerable excitement but, at least initially, appeared to fall short of
expectations. One example is the internet, which despite its widely heralded promise,
remained narrowly focused on information search and content for longer than many
anticipated. Eventually, the internet began its shift to large-scale ecommerce as core
infrastructure emerged to enable an expanded set of use cases. Those innovations
also speak to the ability of key breakthroughs to unlock broader and more
transformative growth across larger ecosystems.
rom the perspective of early 2024, AI appears to find itself in a position surprisingly
F
similar to the early 1990s’ internet. Despite its clear potential to profoundly transform
society, for the moment AI remains largely confined to information and content
creation. But whereas breakthroughs in commercial transactions dramatically altered
how the internet is used, a similar transition to a commercially impactful AI appears
less imminent, for two important reasons. One is that virtually all large language
models (LLMs) powering today’s AI agents are not enabled to handle value or engage
in commercial transactions. In addition, the inherent autonomy in decision-making and
the black-box nature of most AI applications poses a particular challenge for
participation in traditional financial systems. The opacity surrounding ownership and
control over many AI agents means that even if leading AI agents were capable of
conducting financial transactions, it is difficult to imagine them acquiring access to
bank accounts to store value, or to traditional financial markets enabling them to
transact. While these are obvious problems in the near-term, it is not obvious these
issues will be resolved even over intermediate timeframes.
hese barriers are significant, because for AI agents to have any meaningful
T
participation as autonomous actors in the economy—an important step for AI to begin
fulfilling its larger promises—AI agents must first have direct access to, and
independent agency within, some sort of financial value system where they can
transact assets they control.
espite these constraints, there is a possible solution: natively digital, decentralized
D
blockchain networks are ideally suited for acting with the necessary speed to
satisfactorily engage with the rapid nature of AI information processing and
decision-making. One scenario is enabling AI agents to interact with blockchain
networks as a means of holding assets, transferring value, and engaging with broader
financial activities. While not without its own challenges, this approach suggests that
6
igital, decentralized, and community-controlled financial ecosystems might be the
d
most realistic path forward for enabling a wave of AI-focused innovation.
here is clear appeal in using decentralized blockchain networks as the value transfer
T
mechanisms of AI agents. However, this approach also faces a number of significant
difficulties. Notable amongst these are questions concerning secure means for
providing AI agents with access to digital Web3 wallets and the private keys
controlling them. Without these steps, AI agents cannot autonomously control assets
or transfer value. Equally challenging is that AI agents have no native abilities to
navigate and transact within blockchain ecosystems. Addressing these challenges is
critical, however, for if these problems can be addressed, autonomous AI agents could
become dominant sources of transactions, optimizations, queries, and smart contract
deployment across all blockchains.
In such an environment, Wayfinder intends to serve as the primary gateway and set of
rails for AI agents within blockchain environments. In effect, Wayfinder intends to
commoditize the navigation tools and infrastructure enabling interaction with the
networks, bridges, contracts, and applications that define blockchain ecosystems.
Through its novel structure for incentivizing and rewarding the indexing and ranking of
relationships between blockchain nodes, and by establishing new edges and nodes
that become navigational wayfinding paths for the network’s AI agents to utilize while
traversing blockchain environments, Wayfinder will become an essential tool for a new
era. To phrase it another way: while some protocols are making important
contributions to the blockchain space through incentivized intelligence, Wayfinder
proposes to bring important contributions through incentivized AI agent capabilities.
ayfinder’s decentralized, community-governed ecosystem, structured around tools
W
enabling AI agents, as natively digital intelligence, to access and freely transact within
natively digital economies via blockchain ecosystems, digital wallets, and
cryptocurrencies, promises to substantially address the primary challenges currently
limiting the potential of AI agents in blockchain contexts. This is a critical step, for if
one assumes the existence of an interactive on-chain intelligence as embodied in AI
agents, it becomes clear that the expansion of its interactions and capabilities,
particularly in regards to a capacity for digital adaptability, becomes a significant
strategic advantage. In short, this represents what effectively becomes a sort of
unnaturalselection within digital ecosystems, withthe capacity to unlock enormous
new possibilities.
hile Wayfinder provides technical innovations enabling permissionless access for AI
W
agents to freely interact and transact with assets they control, it also represents the
development of a new architecture enabling AI agents to confidently traverse
blockchain environments, in effect moving assets from a wallet they control across a
bridge to a specific application. These agents can analyze markets or other
opportunities upon arrival at a destination, engage in transactions, then return back to
their starting point. While significant for their inherent technological innovations, these
7
fforts also promise to enable a more significant transformation from AI’s current
e
reality of a narrow search and content focus to a broader, more expansive one of
genuinely decentralized commerce. In developing these tools, Wayfinder has taken an
important step towards unlocking the trajectory followed by the internet, and the
potential implications of these innovations may ultimately enable a far wider, and
potentially more transformative, series of impacts.
hese breakthroughs rely upon innovations relative to a number of outstanding
T
challenges. One is the provision that AI agents (referred to as “shells” within
Wayfinder) require the ability to securely execute on-chain transactions, which is a
critical element of the network’s architecture. Another challenge is providing a
framework for shells to easily and accurately access specific smart contracts
facilitating core functions like swapping tokens, bridging across blockchains, and
transferring assets. Essentially, this requires providing a map guiding AI agents across
decentralized ecosystems and from application to application.
ayfinder’s solution to this problem—what the network terms “wayfinding”—has the
W
potential to dramatically improve the accessibility of blockchain ecosystems, and allow
a wide range of users to more easily interact with and navigate blockchain
ecosystems. Just as indexing proved crucial in unlocking the internet’s larger potential,
indexing blockchain smart contracts as wayfinding paths enables AI agents to
seamlessly navigate blockchain networks. Leveraging the impact of these
wayfinding approaches by incorporating bespoke, user-friendly interfaces
powered by a diverse suite of LLMs ensures these tools are accessible to all, and
are not limited to those with advanced coding skills. Wayfinder enables this
accessibility by incorporating LLMs as the command interface, ensuring that AI
agents are able to understand human intentions and execute functions proactively,
based on a generalized set of instructions rather than detailed, step-by-step micro
commands.
mpowering AI agents to control digital wallets and exchange items of value opens
E
a universe of possibilities. For instance, Wayfinding enables AI agents to navigate
blockchains and unlocks accessibility, while incorporating LLMs renders these
tools accessible to a broad, non-technical audience. Collectively, these innovations
also present a novel solution to another significant challenge that has long been at
the center of many discussions and research efforts within the blockchain space.
That question is one of interoperability, and the challenge of moving assets across
chains and between ecosystems, in effect freeing digital assets from the
complexities of interacting with bridges, multiple wallets, and different protocols
and applications. This has long been understood as a factor that has kept the
broader blockchain ecosystem in a fractured state, presenting imposing burdens
for those attempting to move efficiently across different corners of the space.
8
ayfinder, while not presenting a technical solution in terms of a new blockchain,
W
consensus mechanism, or faster bridging solution, nevertheless may represent a
significant step forward—one towards a seamlessly interoperable network of
blockchain ecosystems. By effectively obscuring the complexities of cross-chain
operations on a practical level, Wayfinder allows technical and non-technical users
alike to direct an AI agent to follow a wayfinding path leading from one blockchain
to another, transfer assets, and to return “home,” all with only generalized
instructions. A new world of user-friendly interoperability is enabled. In effect, a
user knowing what they intend to do is able to speak to their shell, and the shell is
able to interpret the user’s intent and complete the transaction. While this
approach does not reduce the inherent technical complexity of blockchain
interoperability, it effectively moves much of this complexity under the hood,
making the challenges of interoperability vastly more manageable for users.
s this paper examines, these innovations are the primary areas of opportunity for
A
Wayfinder. Currently, forecasting where Wayfinder will unlock new possibilities for AI
within the blockchain space remains challenging given the scale of sectors Wayfinder
may profoundly impact. Its effects may ultimately be seen within the gaming world by
materially improving general accessibility within a Web3 context, or perhaps in terms
of trading, financial markets, or NFT collecting. What is clear, however, is that
Wayfinder holds the potential to profoundly disrupt the status quo across a range of
disciplines, effectively to introduce transformations that might eventually be evaluated
as on par with transformative innovations like indexing the internet and rendering it
easily accessible and navigable to millions of users. From this perspective, the
network’s ultimate impact may only become evident over time.
Protocol Overview
ayfinder is an AI agent framework able to power blockchain-based games, with
W
Parallel Studios’Colonyas an initial game incorporatingWayfinder’s technology, as well
as provide generalized on-chain AI capabilities to both developers and non-technical
users. Wayfinder’s innovations in enabling AI agents to operate within blockchain
environments is achieved via the development of what is being termed “wayfinding
paths,” which serve as an instruction set and roadmap for AI agents. This enables
them to navigate blockchain environments with the goal of attaining a unique
endpoint, such as a particular DEX on XYZ chain, a decentralized marketplace, or a
Web3 enabled tool. These transactions might include swapping a particular token for
the most liquid stablecoin available, minting an NFT, or acquiring access to gated
content. Ecosystem participants owning a Wayfinder shell are expected to be able to
pay a small fee in the network’s proposed native token to direct their shell to utilize the
network’s collection of wayfinding paths. These wayfinding paths enable diverse
actions, including traversing blockchain environments and undertaking a wide range of
operations that can include trading, collecting, and minting assets, among others.
While these examples are relatively straightforward, the broader range of potential
9
ctions is effectively unlimited and includes most any activity that currently occurs
a
within blockchain ecosystems.
ayfinder will launch with an initial set of wayfinding paths intended to support access
W
to and interaction with the most popular applications within a wide range of blockchain
ecosystems including, but not limited to, Ethereum, Solana, Base, Cosmos and other
popular blockchain networks defined by the community via governance participation.
On these networks, the initial wayfinding will immediately enable users to interact with
applications and direct their shells to undertake various actions such as trade, dollar
cost average, schedule tasks, establish complex contingent orders reacting to news or
market movements, conduct or participate in NFT drops, facilitate collection, or even
create bespoke and recursively improving yield optimization strategies. Additional
functionality is theoretically almost limitless, but will likely require new exploratory
“wayfinding” work to establish navigational paths to a wider range of networks and
applications than will be available at the launch of Wayfinder. These additional
wayfinding paths are anticipated to be added over time, in an incentivized
crowdsourced effort.
ommunity members will be able to earn Wayfinder’s native token when developing
C
new wayfinding paths and, if their proposed paths are verified as accurate and
accepted into the network’s shared resource library, share in the fees generated by
use of those wayfinding paths. The result of this approach is an ever-expanding
knowledge graph and shared knowledge base that is designed specifically for shell
agents. Importantly, the self-improving, continuously learning nature of shells enables
them to benefit from the activities and knowledge of other shells, as learnings and
experiences are shared across the network. The result is the ability of a shell to
securely navigate any smart contract that other shells within the network have
successfully interacted with in the past. In effect, one agent's success in a contract
interaction means the future ability for all agents to access and interact with that new
function.
ayfinder shells will be available in theoretically unlimited quantities, with the
W
network’s native token expected to be required to mint and acquire control of a shell.
Acquisition and use of a Wayfinder Shell will be accessible to anyone, including those
without coding expertise, as a result of the LLM-based interfaces built into the shells,
eliminating the need to write code to engage with or direct a shell. Using the included
LLM, users can effortlessly craft prompts for their shells regardless of programming
ability. They can also incorporate elements such as wayfinding information, price data,
memories and learnings unique to their shell, and even uploaded content when
directing their shells to pursue specific actions or objectives.
nderstanding the potential of Wayfinder, as well as the challenges it is addressing,
U
the innovations it represents, and the hurdles that remain, requires a broader
consideration of the progress of both AI agents on a general level, and the question of
AI agents within the context of blockchain environments.
10
AI Agent Fundamentals
I agents powered by LLMs like GPT models are revolutionizing complex environment
A
interactions. These agents, trained on extensive datasets, can understand and
generate human-like text, making them versatile in various applications. Importantly,
recent innovations have developed methods for AI agents to incorporate an LLM that
can describe, classify, and store the AI agent’s experiences as memories, giving them
the ability to selectively recall specific memories based on context and objectives. In
effect, this research has added meaningful capacity for agents to plan, learn from past
experiences, and incorporate their own histories into future courses of action.1 Such
research has significantly informed several of the approaches influencing Wayfinder.
Agent Overview
t their core, agents exist in some state with some goal. An agent can observe its
A
state and the surrounding environment, make decisions about how to achieve its goals
given that state, and execute those actions to modify its state. It continues in this way
recursively until its goal is satisfied.
Agent Context and Workflow
gents exist in a specific environment and are generally working to solve an
A
optimization problem of some kind. This can be as simple as a chessboard (with
extremely definitive operations and objectives) for a chess-playing agent, or as
complicated and mission-critical as city streets in the case of an agent driving an
automobile. Each agent has a goal (or goals) to achieve in the space: the chess agent
1
For more on these innovations, see, for instance: Wang, Guanzhi, et al. “VOYAGER: An
pen-Ended Embodied Agent with Large Language Models.” MindeDojo: Voyager.
O
https://voyager.minedojo.org. Accessed 6 Feb. 2024.
11
ants to beat its opponent, the driving agent wants to get from point A to point B
w
without crashing (and while also optimizing for potentially hundreds of other
parameters).
gents have the ability to perceive the space they exist within via data provided to
A
them. In the case of the chess agent, it’s the game board; in the case of the driving
agent, it’s a range of sensors, including GPS, multi-angle video feeds, and potentially
LIDAR, among others. These data are fed to the agent, allowing it to understand the
state of its environment and formulate decisions allowing its goals to be achieved.
he agents themselves digest data and, using some previously learned logic and some
T
reference data or “memory,” will attempt to make a decision that maximizes the
likelihood of it achieving whatever goal(s) it has set out to accomplish. The decision
made by the agent is actioned via an interaction with the environment that modifies it
(or doesn’t, if so chosen by the agent) in some way. This is done by allowing the agent
access to a specific set of tools or interfaces, such as the API to move the pieces on
the chess board, or the steering wheel and brakes of the autonomous car. Once the
interaction is completed, the environment has changed and it is once again observed
by the agent to render a decision—the cycle continues until the objective is achieved.
hile these flows are common for AI agents, and have already achieved meaningful
W
success, a number of challenges exist for efforts to bring these approaches to
blockchain environments.
12
The Blockchain Environment and its Agents
Challenges for AI Agents in Blockchain Contexts
espite the significant opportunities and innovations AI agents are poised to bring
D
when operating within blockchain ecosystems, there are notable challenges and
potential shortcomings in current infrastructures that substantially restrict the present
capacities of AI agents to impact these environments. Considering some of these
challenges helps appreciate the context in which Wayfinder is developing and the
solutions it continues to refine.
Lack of Context
I agents operating in blockchain ecosystems face significant obstacles due to the
A
inherent complexities of smart contract interactions. For a seemingly simple task like
swapping one ERC-20 token for another, an AI agent must navigate multiple steps.
These include locating the addresses of the involved tokens, identifying an
appropriate contract for the swap, and comprehending the smart contract to
determine the necessary functions for execution. This process becomes exponentially
more challenging, and expensive, when considering price comparisons across
decentralized exchanges (DEXes) and verifying the support for specific trading pairs.
o execute these tasks without a contextual framework, the agent would need to
T
engage in extensive data scraping and consume significant resources identifying and
evaluating the range of potential components of a single swap. Many of these
components might have been addressed in previous transactions by other entities,
illustrating the potential wastefulness of endlessly duplicative efforts. This not only
leads to inefficient use of resources but also highlights AI's inability to effectively learn
from and leverage historical data or collective intelligence.
urthermore, these challenges are compounded by the dynamic nature of blockchain
F
ecosystems, where new tokens, smart contracts, and DEXs are continually emerging.
AI agents often lack the intuitive understanding of contractual nuances and the
interconnections within a network of smart contracts, making their navigational and
decision-making processes even more uncertain.
Efficiency
any existing blockchain platforms suffer from slow transaction speeds and high
M
latency. This sluggishness is a significant hurdle for AI agents requiring real-time or
near-real-time data processing and decision-making, particularly in fast-paced
environments such as financial markets or real-time strategy games.
lso of concern is that operating advanced decision engines that possess intuition for
A
the blockchain and the context necessary for understanding smart contracts is costly.
While costs associated with these LLMs can reasonably be expected to decline over
13
t ime, the pace and extent of any potential declines remains uncertain, and this may
continue to be a substantial challenge for AI agents operating within blockchain
environments.
Shortcoming of Existing Tooling
he current landscape for developing blockchain-integrated AI agents is both limited
T
and markedly inefficient. While there are numerous frameworks for creating AI agents
in general applications, the tools for blockchain-specific developments are scarce and
often inadequate. Gaining access to current blockchain data, which is crucial for
optimal agent performance, is often costly, and at other times can be a formidable
challenge at any price. Developers are faced with the task of either constructing
numerous custom integrations to connect agents with blockchain data or settling for
the few available tools, even when they offer limited functionality. This severely
restricts the agents’ potential for creativity and resourcefulness, such as coding
custom Python routines, which are essential for making AI agents truly effective and
versatile in the blockchain domain.
hese issues represent just a few of the major headwinds for attempts to integrate AI
T
agents into blockchain environments. Even this brief consideration helps illustrate
some of the core challenges Wayfinder is addressing, as well as developing an
appreciation of the unique solutions Wayfinder is developing.
Advantages of Blockchain Environments
espite the challenges blockchain environments pose for AI agents, the significant
D
potential for blockchain-focused AI agents unquestionably justifies efforts to address
these challenges, as noted in the following partial list of advantages.
Advantages of AI Agents On-Chain
he integration of AI agents into blockchain environments enables a paradigm shift in
T
how these agents operate within, and interact with, digital spaces. While there are
numerous advantages of such integrations, some of the most significant include:
● E
ase of Ownership:on-chain AI agents enable structureswhere ownership and
control of the agent is determined by possession of a dedicated token—a
simpler structure than needing to establish a legal entity to manage the
ownership of an agent. Instead, the inherent simplicity of a token in a wallet is
sufficient to signal ownership and control of an agent.
● C
omplex Contextual Understanding:Well-trained AIagents are able to grasp
nuanced details and contexts, which is crucial for interpreting ambiguous
information. This is particularly true when agents are properly equipped with
the specialized domain knowledge required to interpret specialized questions.
14
hile context has been a challenge for AI within blockchain environments, this
W
also illustrates the scale of opportunity for those able to address this difficulty.
● A
daptive Learning:By continuously learning from newinputs, agents remain
effective in dynamically changing environments.
● L
anguage Processing:The ability of agents to processand generate natural
language facilitates seamless interaction with humans and other AI systems.
This offers particular opportunities within a blockchain environment, when the
ability to direct an agent to undertake complex transactions can be done
through generalized instructions and without the need to precisely specify the
range of details for all the complex steps that may be involved.
● D
ecision Making and Problem Solving:AI agents arecapable of advanced
decision making and critical thinking that rivals human performance under
certain conditions. Expectations for a continuously expanding range of
conditions within which AI agents outperform are widespread, and if validated,
will soon result in a wide range of situations where AI agents excel.
● I nnovative Solutions:Drawing from a vast knowledgebase, AI agents propose
creative problem-solving approaches. As the internet continues to expand,
these agents will have updates to their context, improving their abilities to craft
novel solutions to queries.
● R
isk Assessment:Agents are capable of efficientlyevaluating risks and
potential outcomes in order to make informed decisions.
● E
thical Considerations:Agentshave shown themselvesto be increasingly
adept at making decisions that consider ethical implications and societal norms.
(While this is generally accurate, Wayfinder does not seek to overly rely on
agents to perform this function in connection with critical activities).
● L
earning Processes:Agents benefit from constant refinementbased on
feedback that enhances their accuracy and relevance (feedback loops). Agents
can apply knowledge from one domain to others for rapid adaptation (transfer
learning). They can also learn alongside other AI systems and humans to
develop comprehensive understanding (collaborative learning).
eyond this list, AI agents partnered with LLMs promise to bring a transformative edge
B
to handling complex scenarios by combining advanced decision-making with iterative
learning. Their careful integration into blockchain ecosystems promises decentralized,
secure, and efficient operations.Within the contextof Wayfinder’s objectives, when AI
agents are enabled with a Web3 wallet and access to the wallet’s private key, the
potential to autonomously conduct financial transactions without needing access to
traditional banking systems is tremendous. This autonomy enables agents to control
15
ssets including NFTs and digital rights, and engage in complex financial activities like
a
trading, minting NFTs, investing, or managing digital assets without human
intervention, further broadening their functional scope. These abilities enhance
operational efficiency and create new possibilities for autonomous economic agents.
Furthermore, when coupled with the aforementioned functionality and advantages of
on-chain agents, this autonomy allows agents to observe the economic impacts of
their actions far faster and more accurately than traditional mock trading and
backtesting approaches, all with a fraction of the development effort of traditional
algorithmic trading methods.
16
Key Elements of Wayfinder
Architectural Overview
ayfinder’s underlying architecture and blockchain navigation mechanisms are
W
proposed to be chain-agnostic tools, intended to serve as foundational rails enabling
AI agents to function across ecosystems. To achieve this goal, Wayfinder’s initial core
knowledge base—including wayfinding paths, as well as additional tools and
instructions —will enable shells to manage wallets and digital assets. It will also allow
them to interact with multiple blockchain networks, including address, fee, and
operational information. Wayfinder will launch with an initial graph of major entities
within the popular blockchain ecosystems, so that Wayfinder AI agents are able to
natively interact with and understand popular blockchain ecosystem smart contracts.
critical part of the network’s architecture and shared knowledge base consists of
A
directions for shells regarding usage of the private keys for the dedicated wallet each
shell will control, a process described in more detail below. In addition, this knowledge
base also provides information ensuring that the LLMs comprising the user interface
understand blockchain ecosystem graphs, and is specifically developed to ensure that
shells excel at incorporating their knowledge of Wayfinder’s ecosystem graph in
responding to user prompts and requests, particularly for instructions to observe and
execute on-chain actions from the shell’s wallet. Of note, although popular LLMs
prevent AI agents from making financial decisions for humans, Wayfinder has
incorporated and modified open-source alternatives in order to create a modified LLM,
enabling functionality that becomes critical to Wayfinder’s vision and objectives. In
addition, and despite the effort to develop these tools, both the data set and the LLMs
will be open-sourced by Wayfinder.
he Wayfinder stack, representing a multi-part framework designed to address the
T
challenges shells face within blockchain ecosystems and the democratization of
blockchain agent technology to allow for a proliferation of AI on major blockchain
networks, is summarized in the following diagram:
17
Wayfinder’s Technical Approach & Key Tools
ayfinder is proposed to be engineered as a chain-agnostic, omni-chain tool that is
W
proposed to reside on the Solana blockchain but aims to seamlessly work across an
expansive set of blockchain ecosystems. Wayfinder is optimized for the high-speed
transactions use cases of Wayfinder Shells, and offers accessible, user-friendly
interfaces intended to simplify the process of engaging with blockchain environments.
By offering easy data accessibility, simplified authentication, and advanced memory
retention mechanisms, as well as a fast and efficient structure for shells to access
wayfinding paths and undertake transactions, the network empowers shells to operate
accurately, efficiently, and autonomously within a range of blockchain ecosystems.
Consensus & Technical Considerations
ayfinder has a primary task of validating the set of wayfinding paths deemed to be
W
safe and correctly specified. To achieve this, Wayfinder will deploy Verification Agents
as central elements of this process, occupying a primary role in evaluating new
wayfinding paths and ensuring the accuracy and safety of both newly proposed
wayfinding routes as well as existing paths already in operation. Initially, a select group
of trusted, community-maintained Verification Agents will work to verify proposed new
18
ayfinding paths, testing for accuracy and effectiveness, and controlling against
w
nefarious pathways. These automated checks will also be supplemented by manual
verification efforts, ensuring the network’s core wayfinding paths have been carefully
evaluated. Pathways that pass these combined manual and automated tests, and are
accepted into the network, will be approved as authorized changes to the graph,
effectively adding new wayfinding paths to the library.
It’s anticipated that, over time, a broader range of community members and
participants will propose additions to the wayfinding library, a process that should
significantly increase the network’s growth and improve the comprehensiveness of its
coverage. Network participants will also be able to operate their own Verification
Agents and contribute to the approval process for newly proposed wayfinding paths.
Private Verification Agents will be able to retain a substantial portion of any bounties
or slashed stakes resulting from successes in identifying faulty wayfinding paths,
triggering the process whereby the staked tokens of wayfinding path developers are
seized.
Purpose-Built LLMs
oundational to Wayfinder is a commitment to leveraging and enhancing open-source
F
LLMs to fit specific use cases. The objective is to create specialized, efficient “expert”
models for different types of shells, increasing efficiency and the effectiveness of
outputs, and optimizing compute resources while reducing costs for users. For
instance, ongoing efforts to fine-tune a compact LLM specifically for EVM function
calls, designed to integrate seamlessly with the wayfinding graph, suggest promising
potential. Specifically, this model is proving to be substantially smaller and more
cost-effective compared to larger, closed-source counterparts, and is expected to
make meaningful impacts upon Wayfinder’s costs for users, overall economic model,
and sustainability.
rucially, the commitment to open-source principles permeating Wayfinder’s work
C
extends beyond the models themselves. The Wayfinder community can propose to
open-source both the fine-tuning of datasets and the weights of these models. This
initiative is aimed at contributing to the community, enabling other developers and
researchers to build upon existing work, innovate, and create more advanced
applications and tools.
User-Friendly Interface for Non-Coders
ayfinder should be developed with a primary goal of making shell creation accessible
W
to everyone, including those without coding expertise. To achieve this, Wayfinder
offers an easy-to-use graphical user interface (GUI) that benefits from intuitive text
templating and systems thinking capabilities. Users can effortlessly craft prompts for
their agents, incorporating various elements such as wayfinding information, price
data, memories, cloning and ownership management, and uploaded content. This
approach enables users to select the underlying model for their agent, determine how
it interacts with other agents, and design its problem-solving strategies. For instance,
19
user could create a news agent that, upon ingesting a headline about a major
a
traditional financial institution entering Web3, decides to buy Bitcoin. This decision is
then seamlessly communicated to a blockchain tool agent for transaction execution. In
addition, a dedicated agent-management UI lets users monitor agent events, set up
stimulation events (e.g., wake up every hour), view and manage the agent’s new and
historical memories, upload any files that the owner would like agents to access, or
even delete stale or otherwise unneeded information.
Comprehensive Integrations for Existing AI Agent Frameworks
o cater to developers seeking to harness the full potential of their AI agents,
T
integrations with popular AI agent frameworks like Langchain, Lamma Index, and
Autogen may be introduced with community governance. These integrations enable
access to the project’s extensive datasets, LLMs, and blockchain tools. To utilize these
features, developers will need an API key, which can be obtained by creating a
Wayfinder account. Additionally, to facilitate seamless transactions and operations
within the blockchain environment, the account must be funded, which is anticipated
to be possible using Wayfinder’s proposed native token as well as a range of other
cryptocurrencies.
Shell Operations and “Explicit Intent”
hile Wayfinder Shells are capable of executing a vast range of operations, and
W
securely and efficiently navigating blockchain ecosystems, it is also important to note
the limits of the capacities and autonomy of Wayfinder Shells. For instance, while a
shell can navigate to a particular DEX to swap two assets within a designated price
window, constrained by an explicitly defined slippage range, shells do not
autonomously manage portfolios in an unbounded manner, transmit value at their own
discretion, or implement novel strategies without the express direction of their owners.
Similar limitations exist regarding the use of Wayfinder Shells in gaming contexts. For
example, while a shell can conduct authorized transactions within a game–such as an
instruction to mine ore and sell the raw materials for the game’s currency–shells do not
autonomously manage portfolios or conduct complex, unstructured operations at their
own initiatives within games.
imilarly, while a shell directed to trade two digital assets is capable of completing the
S
specified trade at the requested venue (or even of choosing a trading venue, given a
set of conditions), the shell would not be capable of completing the requested trade
then deciding for itself to allocate unspent funds without instructions from its owner.
hese examples illustrate that shells are incapable of transacting outside the bounds
T
of the instructions given to them. While instructions to shells can be complex and
incorporate multiple triggers and conditionalities—and actual execution can be equally
complex—effective control of the shell and its transmission of value remains confined
to the shell’s owner. Shells are, in effect, unable to complete any actions that fall
outside of the explicit instructions and permission from the shell’s owner. Reinforcing
20
t hat shells act at the behest and upon the instructions of their owners, prior to a shell
executing a transaction, it will request confirmation from its owner.
Wayfinder, Shells, and Financial Regulation
nother important element of the Wayfinder network is the effort made to ensure the
A
compliance of the network, and in particular Wayfinder Shells, with applicable laws
and financial regulations. Key elements of this include efforts to ensure sanctioned
nations and individuals are not able to access the network. This is accomplished in
part by geolocating individuals interacting with the network, and will be complemented
by incorporating government-published lists of sanctioned digital addresses. In this
regard, shells will also inherit any restrictions or regulatory obligations applicable to
their owners that might be a function of nationality, residence, or other applicable
criteria. In addition, and underpinning these efforts, Wayfinder will coordinate a joint
development, coordination, and review program by several industry-leading
development and auditing industry firms to ensure the Wayfinding platform develops in
a secure way and complies with all general best practices, including all relevant
regulations.
Shell Wayfinding
s noted, blockchain environments present significant challenges for AI agents. The
A
lack of inherent contextual awareness within blockchain data hinders the ability of AI
agents to effectively navigate and interact with complex smart contracts and
protocols. To address this critical limitation, Wayfinder’s shell wayfinding system has
been developed as a sophisticated indexing and knowledge base architecture,
designed to equip shells with a virtual roadmap of blockchain ecosystems. This
enables shells to accurately navigate to the appropriate applications and smart
contracts, even including bridging across ecosystems when appropriate.
hile individual wayfinding paths represent an important innovation, equally important
W
is how the aggregation of individual wayfinding paths evolves into a larger yet detailed
graph that outlines relationships between different blockchain contract addresses,
assets, applications, and wallets. The larger ecosystem graph significantly enhances
the autonomous and transactional capabilities of Wayfinder shells relative to other AI
agents, and provides important new capabilities within blockchain ecosystems.
Particularly important is how this graph enables shells to evaluate alternative paths to
attain specific objectives, ensuring the optimal wayfinding paths will be selected in
each instance. This allows shells to evaluate competing priorities, such as when it
must choose between a wayfinding path that promises lower execution fees and
another that numerous shells have successfully navigated to complete a specific task.
While one user may instruct a shell to prioritize the most well-established and trusted
paths, another may elect to try a new path offering lower costs, but may choose to do
so with a smaller transaction value for an initial test transaction.
21
qually important is that shells are able to leverage this detailed graph through
E
semantic queries, enabling a nearly infinite number of operational tasks, such as
identifying all decentralized exchange pairs involving a specific token, and evaluating
spreads and market depth for each. Once relevant nodes are identified, users are able
to further explore their interconnections using most graph query languages or
semantic queries, enabling a user to develop a deep, structural understanding of
blockchain landscapes. This increasingly comprehensive mapping of blockchain
ecosystems serves as the underpinnings of various AI capabilities, including:
● P
recise Action Execution: Shells empowered by wayfindingcan transition
beyond mere semantic understanding and utilize their knowledge to undertake
precise actions. This includes autonomously navigating NFT marketplaces to
identify relevant addresses, platforms, and collections, and then executing
trades.
● R
etrieval Argument Generation (RAG) Pipelines: Wayfinder’sgraph and its
embedded knowledge link to shells’ RAG pipelines. This increases the ability of
shells to develop contextually relevant interpretations of user queries and
commands, and helps them navigate blockchain ecosystems with greater
autonomy and accuracy.
22
● C
ollaborative Knowledge Expansion: Wayfinder shellsbenefit from a
collectively developed, maintained, and expanded wayfinding system, enabling
them to learn and benefit from the accumulated actions and experiences of the
entire shell population. One example is the way shells will evaluate existing
wayfinding path options when plotting a path for a transaction by assessing the
number of transactions other shells have successfully completed using a
particular wayfinding path. The ability to incorporate the record of a path’s
viability can assist shells in choosing the safest and most established path for a
given objective. Another example is the bounty system: when a shell
encounters a task it cannot complete for lack of an established wayfinding
path, a bounty request is automatically sent to the community. Following an
initial review to ensure the request is valid and the creation of the wayfinding
path would be a significant addition to the ecosystem graph, Wayfinder’s public
bounty list would be updated to include the new path, encouraging other
network participants to establish the correct wayfinding for that objective. This
aids other shells that may attempt the same path in the future.
The Wayfinder Graph
ayfinder’s graph of blockchain ecosystems emerges from a comprehensive
W
combination of unique wayfinding paths and “nodes,” which collectively enable a
broader and more systemic appreciation of the interconnection of different blockchain
ecosystems. These are at the center of Wayfinder’s notion of “smart maps” as
wayfinding paths mapping new smart contracts Wayfinder Shells are able to interact
with. The nodes in the Wayfinder graph will represent:
23
- rotocols: This node encompasses a set of applications, contracts, and assets
P
that are all associated with a specific protocol. Information about the protocol’s
high-level functionalities and applications is essential to understand its overall
capabilities and usage.
- Contracts: These nodes represent the on-chain logicand applications within a
protocol, each identified by a unique address. The contract node details what
the contract does, acting as a critical component in understanding interactions
within the blockchain ecosystem.
- Contract Standards: Nodes like ERC-20, ERC-721, ERC-1155,and SPL libraries
define the standards applicable to contracts and assets. These standards are
crucial, as they dictate functionalities available to a contract or asset,
influencing how they can be utilized or interacted with in the blockchain
environment.
- Assets: These nodes include assets like tokens andNFTs. Essential information
such as the ticker symbol, contract address, and the standard implemented by
the asset is captured, providing a comprehensive overview of each asset’s
characteristics and its role within the blockchain ecosystem.
- Functions: Represented as nodes, these are segmentsof code executable to
perform specific actions on the blockchain. Functions can operate at different
levels, from individual activities to protocol-level operations like executing
swaps on various platforms. These nodes are pivotal for enabling direct
interactions with the blockchain.
- API Functions: These nodes are generally protocol-levelfunctions that provide
information about a protocol. They are crucial for accessing off-chain data that
are heavily used by some protocols.
- Routines: Routines are sequences of functions organized to complete a
particular activity. Each routine node outlines the necessary steps and
functions required to accomplish a specific task within the blockchain
environment using a specified interface. This structured approach facilitates
the execution of complex blockchain operations.
verall, the wayfinding system and concept of smart maps represents a significant
O
advancement in empowering Wayfinder Shells to interact with blockchains efficiently
and autonomously. Wayfinder’s construction of a detailed knowledge base through a
comprehensive ecosystem graph facilitates precise action execution and collaborative
knowledge expansion. By also incorporating enhanced RAG pipelines to improve
interpretation and decision-making, it paves the way for a new era of intelligent
blockchain interaction.
he graph itself may be stored off-chain, facilitating rapid and affordable access to its
T
basic information. Snapshots will be periodically published on-chain for transparency,
and will allow the ecosystem’s general changes and overall evolution to be recorded.
This proposed architecture remains preliminary and may be revised prior to the public
release of the Wayfinder tool.
24
ransforming the Graph into a Dynamic
T
Toolchain
he inclusion of contract functions as individual
T
nodes is more than just a structural
enhancement; it fundamentally transforms the
graph into a dynamic toolchain for shells.
Representing each contract function as a
separate, callable entity allows shells to directly
interact with blockchain contracts. This
structure establishes the graph as an
actionable landscape, where shells are able to
navigate and gather information. It also allows
them to execute blockchain operations directly
through these function nodes.
o effectively handle dynamic data such as prices, transaction volumes, and contract
T
states, information is stored externally. This enhances query efficiency and system
scalability by preventing the graph database from being overwhelmed with
high-frequency, time-sensitive data.
Wayfinding Path Creation and Wayfinding Path Types
central element of the Wayfinder ecosystem is the library of wayfinding paths that
A
enable shells to navigate blockchain environments. The network will launch with a core
set of wayfinding paths established and maintained by the Wayfinder community.
These wayfinding paths will allow shells in the network to immediately navigate core
25
lockchain ecosystems and many of the most popular decentralized applications.
b
Post-launch, and once these wayfinding paths are in use, their ongoing safety and
effectiveness will be monitored by community-maintained Verification Agents, which
will periodically check that existing wayfinding paths continue to operate as promised.
Shells using these wayfinding paths will pay fees in Wayfinder’s proposed native token
for each use, from the wallets they control. These wallets must be funded by the shell
owner. Because developers are rewarded for creating wayfinding paths, both via
capturing bounties and by receiving a portion of fees from other shells utilizing the
wayfinding path, all community stakeholders are incentivized to contribute to a single
shared ecosystem graph and knowledge library. This knowledge library becomes a
shared collective intelligence, allowing shells to securely navigate the expanses of
on-chain worlds. In effect, if one developer or shell has visited a location when
navigating in “public mode,” all shells have visited that location.
s a part of the wayfinding path creation process, paths can be created on testnets
A
for supported networks. These wayfinding paths on testnets will not be part of the
larger Wayfinder Graph during the testnet stage, but will likely be eligible for rewards
from the protocol or application operating on a testnet. Eventually, as testnet
applications migrate to mainnet, testnet wayfinding paths will be quickly transferrable
to mainnet, submitted to the network, and, if successful, approved and added to the
larger Wayfinding Graph. This process should help ensure the larger Wayfinder Graph
remains current and continually incorporates the most recent network upgrades and
emerging protocols.
Seeding the Network’s Initial Wayfinding Paths
In the initial phase of the Wayfinder ecosystem’s configuration, the creation of
wayfinding paths will be meticulously executed by the Wayfinder community for core
destinations related to each blockchain protocol. Documentation pertinent to each
protocol will be thoroughly scraped to extract contract details and necessary
Application Binary Interfaces (ABIs). Subsequently, this data will be carefully filtered
and compressed, utilizing multiple LLMs, to refine and structure the information. This
process ensures the data is optimally organized and formatted, facilitating easy
access and utility for shells interfacing with the respective protocols.
In the subsequent phase, the “Wayfinder Graph Architect” accesses and further
enhances the consolidated information repository. Its primary function is to analyze
and understand the various activities associated with each protocol. It delineates the
required routines for these activities, and programmatically formulates the necessary
contract and API functions for their execution. This comprehensive data, including
details of activities, routines, and functions, is systematically integrated into the
Wayfinder Graph database.
Crowd-Sourced Establishment of New Wayfinding Paths
hile the network will launch with the noted core set of wayfinding paths, the
W
intention is for the network to expand and evolve as community members develop and
26
ubmit new wayfinding paths. When submitting a new wayfinding path for review, the
s
member will be required to stake Wayfinder’s native token as an indication of their
good faith effort and their belief in the efficacy of their proposed path. These paths
will be evaluated by the community’s Verification Agents, as well as human observers
appointed by the community. Any proposed wayfinding paths found to be incorrectly
formatted, or potentially dangerous, will see the proposer’s stake slashed.
hen a community member receives approval for a submitted wayfinding path, that
W
path is processed by the Wayfinder Graph Architect, then added to the network’s
approved library. The path is then part of the shared knowledge base available to all
shells. Any path that is incorporated into the shared library will be available for use
without limitations by all shells within the ecosystem, as long as they are able to pay
the required fee. The developer of the wayfinding path will share with the protocol all
fees generated when shells within the network use their pathway.
hen a member has a wayfinding path approved and added to the shared knowledge
W
library, they are not able to withdraw the tokens staked at the time of submission.
These tokens will be required to remain staked as long as the wayfinding path is
active. This stake will be slashable, and may be used to compensate for any losses
should the path lead to errors or other failures that involve loss of assets. Effectively,
the creator of a wayfinding path has an obligation to ensure their path remains
accurate, even after it has been accepted into the Wayfinder Knowledge Library and
become operational.
hen the token stake associated with a wayfinding path is below a minimum level
W
relative to the average daily transaction value on the path (expected to be around
one-third of daily transaction value), the developer will be required to allow a portion
of the fees they receive from path use to be added to their stake until their staked
token amount has increased to a satisfactory level (expected to be around two-thirds
of average daily transaction value on the path). In addition, Wayfinder will maintain a
list of wayfinding paths where the staked amount is below the required threshold, to
act as a cautionary note for potential users. These warnings will be maintained until a
developer’s stake has increased to the required level.
ltimately, network governance will play an important role in determining the exact
U
details concerning required staking levels and the fees retained for a creator’s stake
when necessary. This governance will likely be involved in evaluating and monitoring
the effectiveness of the ecosystem’s complete set of rules and practices. In addition,
when or if an established wayfinding path has a large associated stake and begins to
see activity on the path decline—a likely scenario if older paths find themselves
displaced by newer paths associated with newer protocols—a developer would see
some amount of their existing stake returned to them, as long as this would keep the
remaining stake above the required threshold.
27
Trusted Participant Incentives
or each participant that creates wayfinding paths that accumulate substantial
F
transaction volumes, the path developer will be eligible for tiered exemptions to the
transaction fee redirections (for example, fees being held back in order to meet
increased minimum stake requirements due to an increase in average daily volume for
a given wayfinding path).
hese reductions could be based on combined transaction volume across all
T
wayfinding paths a creator has developed, as proposed in the following table.
or example, once a wayfinding path creator hits 10,000 valid transactions from all of
F
the paths they authored, 4% of their allocated gross transaction fees would become
exempt from redirection. At 100,000 valid transactions, this would increase to 8%. The
intention is to ensure that successful wayfinding path creators will always receive
some portion of their expected fees, and will not find themselves effectively penalized
for creating successful paths by needing to constantly defer realization of their
expected fees in order to increase their required stake levels.
hese levels and amounts will be subject to modification by token holder governance
T
as the Wayfinder network becomes operational, and as additional data on user
behavior and wayfinding path usage becomes available.
Wayfinding Path Background and Security
hile the initial review of all prospective wayfinding paths and the slashable creator
W
stake substantially reduce the risk of errors and failures, Wayfinder is keenly aware of
the importance of path safety. There are several measures that could be pursued in its
efforts to maintain a constant focus on the safety and security of wayfinding paths
even beyond their launch. These may include qualitative and quantitative tools made
by the community open source developers that enable the Wayfinder community to
evaluate the relative safety of different wayfinding paths.
nother mitigation could be maintaining publicly available, community validated data
A
regarding a path’s creator, in the interest of allowing potential users to better evaluate
28
path’s safety and trustworthiness. These efforts could include both formal and
a
informal elements. For instance, to the extent specific and meaningful information
about the path’s creator is known, the identifier associated with the creator will be
augmented to indicate that the creator is a “Verified Wayfinder Contributor.” As a
hypothetical example, if Wayfinder has manually confirmed that a path functioning as a
bridge to Solana was published by a member of the Solana Foundation, a Verified
Wayfinder Contributor icon would be displayed next to the pathway. If, in this
hypothetical example, the Solana Foundation were to indicate a commitment to
maintaining the path’s secure status, an additional identifying badge could be
displayed.
In cases where the creator of a wayfinding path is an unknown, pseudonymous
contributor, general background and contextual information is still expected to be
available to the extent possible. In this case, information would likely be limited to
quantitative information, including the total number of wayfinding paths the creator
has established, and aggregate numbers for total transactions and value transacted on
the creator’s paths. Additional information, such as the amount of slashable stake
associated with the path, as well as any disputes or failures associated with the path,
could also be available.
ach individual wayfinding path will be accompanied by key data providing
E
prospective users with a variety of metrics they can use to evaluate the potential
safety and efficiency of the path. The selection of relevant data is anticipated to
evolve over time, but is expected to initially include the number of successful uses, the
number of unsuccessful uses, the notional volume associated with successful and
unsuccessful uses, fees associated with use of the path, as well as some information
on the temporal clustering of path use—for instance, if a formerly high-volume path
had not been used in weeks. In cases where errors and failed transactions have been
associated with a path, available data will allow prospective users to evaluate the
relative severity of any such errors, as well as their frequency. Also, the amount of
staked tokens associated with each wayfinding path will always be available. Note that
all provided metrics and security measures are subject to change as new information
becomes available, as the ecosystem evolves, and in response to the collective will
and feedback of the Wayfinder community.
eyond initial wayfinding path review, creator stake, and metadata regarding both
B
creator and path usage, an additional protective mechanism may also be deployed.
This last-mile protection involves the ability for a user to engage one of the
ecosystem’s Verification Agents to navigate a wayfinding path prior to the user
attempting to use the path themself. For a small fee in Wayfinder’s native token, a user
can direct a Verification Agent to navigate a path and attempt a test transaction with
minimal token value. This process allows a user to observe the functioning of any
wayfinding path, while avoiding the risk of loss or falling afoul of an exploit.
Collectively, these mechanisms offer Wayfinder users a variety of measures to
evaluate a path, and should substantially reduce the overall risk of capital loss.
29
Establishing Private Wayfinding Paths
hile the shared library of publicly available wayfinding paths is at the heart of
W
Wayfinder, it is also possible for individuals to create “private” wayfinding paths. These
private paths will be accessible to only a limited number of shells that purchase
licenses for use, or to shells that pay higher fees, which developers will be able to
establish based on their determination of the value the path will have for users. There
are at least two distinct ways in which access to private wayfinding paths can be
structured.
egardless of how access to private wayfinding paths is structured, in either case the
R
creator of the path will submit to Wayfinder their wayfinding path in the ordinary
fashion, including a general description of the path and its intended functionality. The
creator would stake Wayfinder’s native token as for a regular submission, and will
indicate that the path is not intended to be a shared resource. As a private path, the
path will not undergo verification by Verification Agents in the same manner as
proposed public pathways. As such, the path can be added to the shared resource
library, but segmented into a “private” section. The path will not have its detailed
description generally available to all shells, and it will be clearly noted that the
pathway has not been verified.
ermission to utilize private networks could be established in two ways. One is for the
P
pathway to require a higher fee in Wayfinder’s native token for shells wishing to use it.
In such cases, shells would pay a per-use fee higher than those required by public
pathways. The other is for the creator of the pathway to offer a fixed number of
licenses to use the pathway available. These licenses could be priced to allow the
purchaser to use the pathway without additional usage fees, but could also be
structured as a license fee that still requires additional fees, paid in Wayfinder’s native
token as with all other wayfinder fees, on a per-use basis. The choice will be at the
discretion of the pathway developer. Purchasers of a license would have a right to use
the pathway for the defined term of the license, either perpetual or fixed-term, and
would be free to resell their license.
Wayfinding Bounty Program
hen a shell encounters a request it is unable to fulfill—for example, proceed to ABC
W
exchange on XYZ protocols and purchase token EFG—the network will send a bounty
request to Wayfinder which, after a review to ensure the request is deemed worth
adding to the Wayfinder Graph, will post a bounty to a bounty list. Wayfinder will
maintain this bounty list and its contents will be publicly viewable. This will encourage
others on the network to develop that wayfinding route in order to capture the
available bounty and receive a portion of future fees earned as that route is used.
Bounties will be rewarded in Wayfinder’s proposed native token, and will be funded by
Wayfinder and its portion of fees from shells using wayfinding paths. Bounties will be
disbursed once three criteria are met: a wayfinding path must be proposed in
response to a bounty, the path must pass the standard verification checks and, once it
30
is approved, the path must be added to the network’s library. Human operators can
also manually fund additional bounties for undiscovered paths, with these additional
bounties added to the Wayfinder-funded bounties that are automatically generated
when undiscovered wayfinding paths are identified.
hen an individual decides to identify and submit a new wayfinding path, they will be
W
required to provide a stake in Wayfinder’s native token. If the wayfinding path is
verified and approved by the community—using both AI Verification Agents and
humans completing manual evaluations—the path’s creator (or owner) will be required
to maintain their token stake as long as the wayfinding path remains active. As with
regular wayfinding path submissions, staked tokens are slashable should the path be
shown to be invalid. Given that the path owner’s stake remains at risk while the
wayfinding path remains active, the owner must continually ensure that the pathway
remains correct. It's important to note that the policies, procedures, and mechanisms
described herein are subject to modification as Wayfinder’s ecosystem evolves,
technological advancements emerge, and based on community feedback.
31
Overview of Wayfinder Shells
I agents are significant in that they are productive. AI agents are also highly
A
customizable: they are able to be trained on a number of specialized tasks, priorities,
and permissions, and to autonomously interact with specific contracts, all without
human intervention beyond initial training.
ithin the context of Wayfinder, when individuals spend Wayfinder’s native token to
W
create a new shell, they will be able to access the network’s shared repository of
knowledge available to all shells. Shells also have access to the shared resource of
validated wayfinding paths, allowing access to different blockchain ecosystems,
applications, and smart contracts. At their initial moment of creation, shells are
effectively unspecialized vessels ready to be developed in specific directions and with
particular areas of expertise, all depending upon the owner’s intentions for the shell.
Beyond these common resources, however, owners will be able to “train” their agent
by allowing it to ingest files—for example, PDFs, text files, code, images, or
audio—that will permit the shell to improve its knowledge of a particular domain.
Examples could be providing a shell with third-party reports on DeFI protocols, or
feeding it a number of whitepapers from a series of related projects that will enable
the shell to monitor the relative performance of these protocols and trade some
basket of tokens in light of those performances. In each case, the user’s shell would
access its unique learnings and become differentiated from other shells within the
network, even as it retains access to the shared set of tools and resources.
Shells benefit from the following functionality:
● M
emory Retention and Transferability:Wayfinder includesa sophisticated
memory retention mechanism for shells. Shells can write and read memories, a
feature augmented by the network’s proprietary RAG pipeline. This mechanism
ensures that shells have a continuous, evolving, and editable2 memory,
enhancing their learning and decision-making processes. Shells can also have
their memories, experiences, and learnings cloned by their owner, who could
proceed to share the shell’s memories and knowledge with other entities who
could incorporate that data into their own shell, effectively leveraging the
efficiencies of a highly trained machine worker in additional contexts.
● B
lockchain Payload Awareness:The integration of blockchaindata into shell
operations enables more informed and timely decision-making. Shells can
utilize the comprehensive data interfaces and tooling available through
Wayfinder’s shared knowledge base to make rapid strategic decisions.
2
While the software doesn’t directly allow users to “tweak” or adjust the specifics of an individual
emory, users can permanently save and delete specific memories in order to curate shell knowledge
m
and capabilities.
32
● E
fficient Authentication:Shells can be authenticated efficiently within the
blockchain ecosystem. Wayfinder allows applications on the allowlist to control
shells, which can be enabled and refined on a per-shell basis, allowing secure
and controlled access to the different shell functionalities within discrete
settings. This capability allows for further customization of shells, and supports
the specialization of shells relative to particular tasks for which their
performance can be optimized. One context in which this can be particularly
useful is relative to the use of assets in games, where this feature can allow for
efficient gameplay and limited degrees of automation for routine or repeated
avatar actions.
Wayfinder Shells and Private Key Management
core element of Wayfinder’s solution to the challenge of enabling AI agents to
A
navigate and transact within blockchain environments is the mechanism by which
Wayfinder Shells are able to control Web3 wallets and request signatures from
associated private keys in a secure manner. Given the unquestionable importance of
this topic, additional consideration is justified in determining the role human owners
play in transactions executed by Wayfinder Shells and how the private keys of shells
are managed.
s previously described, the process of creating a Wayfinder Shell involves a network
A
participant spending Wayfinder’s native token to acquire ownership of a shell. The
owner signals to the Wayfinder network that they are the rightful owner of the shell,
and have the sole right to interact with, influence, and direct the shell.
nce the owner signals their activation of the shell, the shell is connected to its
O
memory database, its associated LLM or LLMs, and its private key for signing
transactions and messages. Ensuring that private keys are accessed only by their
assigned shell is one of the most critical security components within Wayfinder. At any
time, an owner could potentially transfer their shell’s ownership to another network
participant.
It is anticipated that future versions of Wayfinder may introduce new, fully
decentralized mechanisms for proof of shell ownership, transfering of shells, key
management (including multiple keys per shell), and transaction signing and approval.
These updates are expected to impact the network by ensuring that shells will be able
to operate with less human oversight in the future and subject to community
development and governance. One or more workstreams will ultimately be established
as the network’s standard approach to shell security and private key management. Any
eventual decisions will be shared publicly.
Shells and Wayfinding Path Discovery
o illustrate the wayfinding path discovery process, take an example of a human
T
operator instructing their shell to acquire a fungible blockchain asset, XYZ. The
operator submits a plain text request such as, “Go acquire an NFT of medium rarity
33
f rom collection XYZ if one is available at 10% below the 90-day average price for that
quality of asset. Provide me with the anticipated overall cost basis prior to execution.”
ith this request having been issued, the shell could evaluate all venues offering XYZ
W
trading, and provide a list of any potential assets meeting the designated criteria.
Estimating the total costs associated with each execution option, and the costs
corresponding with any required interim trading activities, the shell could simply
execute the transaction, or return to ask for the user’s guidance on the optimal venue
for the transaction, depending upon its instructions. If asked to return for confirmation,
the shell could provide the operator with a table illustrating the total costs associated
with each execution method; the shell could also note which execution venues are
currently accessible using established wayfinding pathways within Wayfinder’s shared
repository of routes, and which options would require the creation of new wayfinding
pathways. Considering this information, the operator could evaluate whether it deems
any of the accessible execution options to be satisfactory.
hen an operator chooses to execute a transaction, they will need to ensure their
W
shell has sufficient Wayfinder tokens to fund the compute costs associated with the
underlying LLM usage, as well as other fees the shell may incur, including the use of
existing wayfinding paths. In addition, while the shell could theoretically liquidate
tokens it holds in order to fund the purchase of the desired asset (XYZ), it would likely
be more practical to fund the agent with one or more currencies that are commonly
used as trading pairs, so as to reduce the amount of intermediate transactions and
fees that are required. Additional currencies will also be required for paying gas fees
as shells transact on different networks.
ver time, and as network and wayfinding path usage grows, shells will be able to
O
incorporate the volume of transactions of other shells over a wayfinding path into their
decision-making process. This knowledge can aid shells in evaluating different
potential pathways capable of executing a transaction. Shells will also be able to
review the specific actions (or lack thereof) of the system’s Verification Agents that
test new wayfinding paths on behalf of the network. The evaluation of wayfinding
paths can have a temporal component as well, such as in the case of a wayfinding
path that has seen considerable historic traffic but has not been used or tested by a
Verification Agent recently. A shell might elect to follow a more recently used
wayfinding path in an effort to remove potential risk.
34
Wayfinding Path Risks
here are two primary sources of risk that exist when utilizing existing wayfinding
T
paths:
● T
ranslation Risk. This results from the possibilityof an LLM misinterpreting or
otherwise incorporating incorrect specifications of appropriate wayfinding
paths in response to a user request via the GUI. As with all such activities, there
is always some potential for the translation activity to lead to faulty
interpretations. This risk can be substantially mitigated by the pre-execution
process and the underlying wayfinding mechanisms. Prior to execution, for
example, shell owners can review confidence interval statistics illustrating the
relative conviction in the correctness of the interpretation of the request, and of
the suitability of the proposed wayfinding path. In addition, the shell can be
instructed to request explicit confirmation from its owner regarding the
proposed contract address and intended interaction, thereby providing the
opportunity for a final human review.
● N
efarious Pathway Risk. The second risk concerns inadvertentusage of a
nefarious pathway. In this instance, an attacker could have taken steps to
obfuscate the metadata of the malicious contract. It’s possible that even after a
cursory manual review by the human user, the path may appear to be without
issues. The actual usage of this nefarious pathway could result in a loss,
potentially including all value held within the shell. This risk can be mitigated in
several ways, beginning with a user taking additional time to review the results
of the pre-execution process (and the associated confidence interval and
35
xplicit information regarding the number of previous successful uses of the
e
path). In addition, the underlying mechanism—whereby wayfinding path owners
continually have a token stake at risk associated with each specific wayfinding
path—provides meaningful incremental protection, given that the owner’s stake
would be slashed in the case of an attack. And, as previously explored,
Wayfinder may maintain a list of wayfinding paths where the associated stake is
abnormally low relative to the path’s transaction value—an indicator which
could be seen as a sign of a potentially risky wayfinding path, and may
influence some users to reduce their transaction sizes or find alternative paths.
hile these risk mitigation techniques combine to provide users with reasonable
W
protections, it should be noted that certain attacks (particularly in cases where users
are deploying their shells across uncharted pathways) may be difficult to protect
against in cases where prospective users do not review the proposed contract
address and interactions in detail. Given this identifiable risk, however, it is likely that
Wayfinder and its developer community (potentially via the governance process), will
work to develop additional automated warnings for users in cases where the
wayfinding paths do not have a reasonably high volume of successful transactions or
have been inactive for some time. This would allow users to test such wayfinding
paths using modest funds, in order to limit their exposure. Users will also be able to
reward Wayfinder Verification Agents to test a path prior to their own execution of a
transaction on the same path, which is anticipated to prove a useful addition to the
network’s safety procedures.
Catastrophic Planning Failures
In the blockchain arena, AI agents, driven by advanced LLMs, are often engaged in
complex, multi-step operations initiated by simple prompts. These operations, which
can range from token exchanges to NFT acquisitions, involve intricate processes
including deep token analysis, navigating diverse blockchain marketplaces, initiating
and precisely configuring blockchain transactions, and meticulously verifying
post-transaction states, such as checking wallet balances for confirmation.
he technical sophistication required for these tasks is immense, and even the most
T
sophisticated LLMs are not immune to evaluation or planning errors. Mistakes in any
step of a blockchain transaction, whether due to misjudgment in token valuation or a
lapse in executing transaction protocols, can lead to incomplete or faulty processes.
The risks associated with incomplete transactions in the blockchain space are
substantial, extending beyond financial implications to include security vulnerabilities
and potential integrity issues of the blockchain ledger itself. Such errors risk not only
undermining the transaction in question, but can also have cascading effects, eroding
trust in the system and possibly affecting subsequent transactions. This scenario
underscores the imperative for continual refinement and enhancement of AI
capabilities in blockchain management, ensuring they possess not only transactional
36
cumen but also robust error detection and recovery mechanisms to handle the
a
dynamic, high-stakes nature of blockchain operations.
ayfinder architecture will be designed to address these concerns in a number of
W
ways, with the goal of overcoming the current limitations and enhancing the
capabilities of AI agents operating within its ecosystem. These solutions focus on four
key areas: enhanced environment, improved perception, empowered agents, and
interaction enhancement.These factors represent the values and concerns that
permeate Wayfinder’s architectural agendas and operating structures.
37
Creating and Managing AI Agents
primary element of Wayfinder is the creation and deployment of self-improving AI
A
agents optimized for use cases within the unbounded world of Web3 ecosystems.
This section delves into the technical underpinnings of shell creation and
management, allowing users to wield these intelligent entities as tools for navigating
and transacting within arbitrarily complex on-chain environments.
hells are created with full access to and knowledge of Wayfinder’s infrastructure,
S
shared knowledge base, and wayfinding pathways. These can be supplemented over
time by the incorporation of specific learning and training, as well as agent-specific
histories, comprehensive blockchain data, and real-time market conditions.
Wayfinder’s RAG pipeline seamlessly integrates new experiences and specific
contextual information into the existing knowledge base, continuously refining the
shell’s understanding and decision-making abilities.
38
Genesis of a Shell
1. S
hell Minting: The process of creating a shell beginswith a user spending
Wayfinder’s native token to create a new shell. Upon confirmation of the shell’s
creation, the owner is granted exclusive ownership and control over the AI
39
gent. All fees associated with creating a shell are passed to the Wayfinder
a
protocol.
2. S
kill Acquisition: Each shell comes equipped withawareness of the network’s
shared knowledge base. Beyond that common resource, shells represent
malleable entities, and owners can develop shells for specialized knowledge of
particular domains or tasks by supplying them with appropriate information,
such as specialized reports or data, or via the network’s shell training modules
that will allow for specialization of Wayfinder Shells. Each training module will
have its own Wayfinder’s native token, reflecting the complexity and potential
power of the skills it imparts.
3. S
eeding an Identity: Beyond technical acumen, imbuing a shell with a unique
personality is possible. Defining core traits, goals, and even backstories through
the “seed identity” settings personalizes the shell, influencing its
decision-making processes and shaping its interactions with the world. While
shells will accumulate memories of their actions, and over time see their actions
influenced by their growing bank of experiences, owners can also “pin” certain
memories so that they remain central to a shell’s experience set. This prevents
certain memories from diminishing in influence as they age and become
increasingly smaller parts of a shell’s overall memory bank.
40
Memory and Advanced Features
Generation and Customization of Memories
hells leverage a robust internal memory system to learn and evolve beyond their
S
immediate function calls. This system comprises three key components:
1. C onversation Memories: Captured from user interactions,these memories
provide shells with personal context and understanding of its owner’s intent.
2. Tool Memories (Short-Term): Generated from temporarytools deployed by the
shell to solve specific problems, these memories provide immediate context for
ongoing tasks.
3. Long-Term Contextual Memories: Built from the shell’saccumulated
experiences, including interactions with humans, other shells, and function call
outcomes, these memories underpin a holistic understanding of the world.
sers actively participate in shaping shell memory through different forms of
U
customization. These methods provide fine-grained control over what gets saved and
how, including:
● M
emory Curation: By pinning specific memories, userscan ensure they remain
prioritized even as the shell accumulates new experiences. This is crucial for
preserving critical knowledge and preventing shell expertise from being
unintentionally overridden.
41
● M emory Deletion: Users can also choose to delete unwanted memories,
further refining the shell’s knowledge base and optimizing its learning process.
● Memory Expansion: Given the limited default memory capacity, users can pay
a fee in Wayfinder’s native token to expand their shell’s storage capabilities,
accommodating a more robust and comprehensive record of historical
experiences.
● Memory Wiping: For a more drastic clean slate, userscan initiate a complete
memory wipe using Wayfinder’s token, resetting the shell to its initial state.
earning from past experiences stored in memories, shells escape the trap of
L
repetitive loops and are able to incorporate insights from prior experiences into the
planning of future actions. This continuously accumulating archive of experiences is
ever-evolving in its weighting as older experiences are gradually supplanted in
importance by newer ones. As a result, a single shell’s behaviors and responses evolve
over time, as new experiences and the lessons drawn from them accumulate.
Meanwhile, memory pinning ensures that while shells are continuously learning,
owners can still shape the manner in which shells recall relevant memories at the
appropriate moment, allowing these agents to be highly customizable.
Enhancing Context with External Data
eyond inherent memories, shells can access and leverage user-provided data for
B
enhanced contextual understanding. Potential data sources include documents,
photos, videos, and, eventually, even application controls—essentially any saved
content deemed relevant to the shell’s tasks. For example, an owner training a shell to
predict wayfinding patterns using time series data could provide supplemental
documentation on machine learning techniques like XGBoost.
This empowers the shell to:
● E fficiently Acquire Knowledge: By learning from curateddata, shells can be
rapidly trained to develop specific domain or conceptual expertise, allowing
them to more easily grasp complex concepts and implement or develop
necessary code for specific tasks.
● Perform Tailored Functions: With access to relevantcontext, shells can be
trained for specialized roles like order management or customer service,
drawing upon specific training data to provide human-like recommendations
and services.
hrough seamless knowledge integration and user-driven memory customization,
T
shells evolve into flexible agents capable of learning, adapting, and collaborating on
complex tasks.
42
43
Wayfinder’s Native Token
ayfinder is expected to be fueled by a dedicated token that, if authorized by the
W
community, would likely serve a number of crucial roles within the ecosystem. These
may include functioning as the currency for acquiring shells, securing access to
wayfinding, operating as a general utility token, serving as a gas token and means of
payment, and use as a governance vehicle. This paper proposes for this token,
tentatively called PROMPT, to have a maximum authorized supply of 1 billion tokens.
However, this will be subject to a community governance process that will confirm the
final implementation details.
he Wayfinder Foundation acknowledges the importance of growth of the protocol
T
and will strive to work with popular blockchains and decentralized foundations to help
bootstrap momentum for protocol growth. As such, the Wayfinder Foundation
anticipates supporting existing contributing communities such as Parallel Colony and
the Echelon PRIME Foundation by suggesting to the Wayfinder community that 40% of
the supply of Wayfinder’s native token be distributed to these ecosystems and
potentially other popular blockchains and decentralized network protocols. If adopted,
this governance proposal would see the Wayfinder native tokens distributed to token
holders of ecosystem partners who Cache their tokens on the Wayfinder protocol over
a three-year period. This proposed arrangement is subject to ratification by
community governance via a unique governance process that will begin prior to the
minting or distribution of any potential Wayfinder token.
* Subject to community approval
44
Reference: Command Prompt Key NFT (See table above - Stage 1)
45
Potential Token Supply and Distribution
he proposed Wayfinder native token is proposed to launch with a maximum
T
authorized supply of 1 billion tokens, subject to the completion of the governance
votes. The initial token distribution is anticipated to resemble the following:
*Note that the investor token allocation pool amount is not subject to community governance.
Potential Genesis Allocation
istributions to Holders of Ecosystem Tokens
D
45% of the native Wayfinder token supply is anticipated to be distributed to holders of
ecosystem tokens who cache their tokens within Wayfinder. This distribution, if
adopted, is expected to occur monthly over a three-year period to ecosystem holders
who continue to cache their tokens.
46
ayfinding Rewards
W
It is anticipated that 5% of the overall PROMPT supply will be allocated to support
wayfinding activities. If approved, this allocation will fund bounties to incentivize the
discovery and creation of yet-to-be opened wayfinding paths. In addition, the
allocation will also support rewards issuance to wayfinding path owners as a function
of shell traffic. The Wayfinding Rewards pool is anticipated to be sufficient to fund
bounties and rewards for an initial multi-year period. If or when the pool is exhausted,
additional funding needs will be addressed via the network governance processes.
Cumulative Supply Over Time
47
Primary Token Functions
If the community governance process approves the creation of a native Wayfinder
token, the token could power a number of use cases within the Wayfinder ecosystem,
including:
● Spending Wayfinder Tokens to Purchase an AI Shell
a. Acquiring an AI shell is the first step to participating in Wayfinder. Users
looking to create a shell will utilize Wayfinder’s native token for these
purchases.
● Spending Wayfinder Tokens Within Wayfinder
a. AI shells within Wayfinder will spend Wayfinder tokens as they engage
with both the network and other shells.
b. Wayfinder tokens will likely serve as gas within the network, with gas
fees being shared between the developers of wayfinding paths a shell
interacts with and the protocol.
c. Wayfinder tokens could also function as the exclusive currency used in
connection with the sale of private wayfinding paths.
d. Shell owners will be able to spend Wayfinder tokens to ensure that
specific learnings of their shells will remain private and not become part
of the shared knowledge base available to all shells.
e. Shell owners will be able to pay a fee in Wayfinder tokens to expand their
shell’s storage capabilities.
f. Shell owners will be able to spend Wayfinder tokens to pay for full or
partial resetting of a shell’s memory.
g. Shell owners will be able to spend Wayfinder tokens to fund a
Verification Agent to navigate a wayfinding path prior to the shell
attempting to navigate the path itself. This is expected to be used by
community stakeholders prior to executing a high-value transaction, or
when a seemingly sound path has not been used in some time.
● Staking Wayfinder Tokens to Establish a Wayfinding Path
a. Users stake Wayfinder tokens to propose a wayfinding path.
b. Staked tokens used to propose a path will be required to remain staked
as long as the wayfinding path remains active.
c. Staked tokens are slashable should path routing be shown to be
incorrect.
d. The risk of being slashed at some future date, as long as the wayfinding
remains active, effectively requires path owners to continually ensure
that their pathway remains correct.
● Governance
a. Wayfinder tokens would serve as a governance vehicle within Wayfinder,
allowing token holders to oversee the future direction of the network.
48
Potential Ways to Earn Wayfinder Native Tokens
● E
arning Wayfinder Native Tokens as a Percentage of Throughput Fees on a
Wayfinding Path:
○ Individuals who submit a wayfinding path that is approved by the
protocol’s Verification Agents will be eligible to earn a percentage of gas
fees paid by shells as they utilize that path.
○ Individuals creating “private” wayfinding paths will be able to set their
own rates for a license or fee for use of their path. In such instances, the
protocol will retain 10% of total license and fee revenue generated by the
wayfinding path.
● Earning Wayfinder Native Tokens for Fulfilling a Bounty:
○ Individuals who develop a new wayfinding path in response to a bounty
will be eligible to receive any tokens that have been allocated towards
that path. As with the standard process, anyone submitting a wayfinding
path will need to stake tokens in order to propose their solution. The
tokens will remain staked as long as the wayfinding path remains in
operation. The wayfinding path will not be eligible to earn rewards from
community use until the protocol’s Verification Agents have evaluated
the proposed path and deemed it safe and accurate. As with normal
wayfinding submissions, incorrect or malicious wayfinding will see the
staked tokens slashed.
● E
arning Wayfinder Native Tokens by Capturing Bounties When Operating a
Verification Agent:
○ Verification Agents within Wayfinder will be operated by community
stakeholders. Creators of Verification Agents will be able to capture
bounties and recover a portion of slashed tokens when their agent
identifies a broken wayfinding path or other error within the network.
● A
dditional Programsfor community stakeholders toearn Wayfinder native
tokens will continue to be evaluated
○ The token holder governance process is expected to modify Wayfinder
native token earnings and distribution methods in the future.
49
Wayfinder’s Initial Use Cases
he innovations within Wayfinder lend themselves to a wide range of implementations
T
and use cases. The following sections describe several of these use cases.
Wayfinder’s Substrate for Provably Scarce Generated Resources
ayfinder has a larger applicability for gaming genres. One example is autonomous
W
world ecosystems, where Wayfinder provides a framework enabling a generative
substrate of scarce resources. This substrate is a fundamental building block for
ensuring consistency in generative assets within a simulation.
n example is the case of a miner digging for a material that has been generativity
A
determined to exist within a simulation game: the probability of pulling the item from
the ground—effectively, the item’s rate of inflation—can be governed by Wayfinder’s
tooling. This enables a consistent, verifiable baseline of scarcity across all assets
within a simulation or other game.
Application-Agnostic AI Agents
ayfinder’s user interface is a powerful tool that can be harnessed to transform a
W
generic shell into an expert in a specific domain. This enables users to provide
Wayfinder Shells with focused educational and contextual material in many forms,
allowing for easy customization and specialization.
his capability can be illustrated with an example of a firm providing an agent with the
T
company’s customer service documentation and transcripts of prior communications,
enabling an agent to rapidly become a subject-matter expert capable of handling
various customer support inquiries. LLMs have already had considerable success in
improving the effectiveness and efficiency of help-center agents, and Wayfinder offers
to improve upon those capacities in a broad sense, particularly in the context of
incorporating Web3 and blockchain-based information.
nother example is a user teaching a Wayfinder shell to analyze new smart contracts
A
and filter for appealing new gaming assets based on specific criteria. This agent could
be taught to emphasize a number of variables related to token contracts, and could be
allowed to develop their own methodologies, enabling them to locate tokens they
deem promising. They could also be directed to employ a hybrid method that is
broadly applicable to all games of a specific genre, such as trading card games. A shell
could even be trained to focus on a specific Web3 game with multiple contracts and
assets, in order to develop analytical tools and insights for discerning game trends and
developing novel strategies.
NFT Minting Assistant
user might customize a Wayfinder Shell by training it to become a NFT-minting
A
expert. Such capabilities could be directed towards a professional application—such
50
s an artist who regularly mints NFT collections intended to be sold by auction—or a
a
retail application, such as users wanting to mint small collections of images with
personal value to distribute to friends. In each case, a shell could be trained to easily
ingest the media intended to populate the NFT collection, determine the optimal
format for minting the media in question, evaluate the relative attractiveness of
blockchain ecosystems and minting applications, and finally oversee the mint. Once
the NFTs are minted, the shell could, for a professional, transfer them to the
designated wallet prior to a sale or auction. For a retail user, the shell could oversee
distribution of the minted NFTs to the intended recipients and their own wallets.
dditional use cases are virtually limitless, but these examples provide some insight
A
for directions that users might pursue in the future.
Smart Trading Bot
nother tool a user might create using Wayfinder’s interface is a Smart Trading Bot for
A
digital asset trading and portfolio management. If a user chooses to provide their shell
with didactic materials covering topics such as portfolio management, asset allocation,
and arbitrage trading, the user would then be able to deploy the shell for a variety of
trading and investment management strategies, opportunistically selected based on
user-specific risk tolerances and investment objectives.
t a basic level, a simple dollar cost averaging strategy, whereby the shell purchases a
A
constant amount of a single digital asset or a specified basket over a set period of
time, could be easily deployed using Wayfinder’s tools. More complex applications are
also possible: for example, where a shell undertakes to mirror specific time-delineated
volume weighted average price targets, which requires the shell to have greater
discretion as to when it executes its purchases based on its expectation of future price
performance.
51
Governance
Governance Overview
overnance has three distinct aspects within Wayfinder. One concerns the
G
well-established practice of token holders voting to govern the broader network, as is
common within the crypto ecosystem. Some matters such as ecosystem direction,
staking and fee levels, elections, and reward parameters are representative of the
scope in this context. An important part of the governance process is also expected to
be an ongoing community-led effort to monitor and revise the methods by which
wayfinding paths are evaluated and ranked. Given the anticipated proliferation of
alternative wayfinding paths, the potential range of embedded options for fees, speed,
and security, and the continuous evolution that blockchain ecosystems continue to
see, this is anticipated to become a meaningful aspect of ensuring the network’s
continued safe and efficient operation.
second distinguishing feature of Wayfinder’s governance model is how the network
A
enables AI agents to participate in the governance of their own network, alongside
human stakeholders, without making an active distinction between the two groups.
Because governance is available to all token holders, and because AI shells will be
holding Wayfinder’s proposed native token in order to transact on the network, the
network will allow AI shells both to vote and even to submit governance proposals,
should an AI shell desire to do so. While the decision to allow Wayfinder Shells to
participate in the governance of their own ecosystem is considered to be a core
proposition of Wayfinder, shells are expected to be limited to an aggregate total voting
weight of approximately 20%, given the potential sybil risk that unchecked shell
participation could pose to the network’s stability.
nother distinguishing element of Wayfinder’s governance approach is how it will
A
include an “AI Constitution” representing the highest level of governance over the
behavior of AI shells on the network. This constitution will establish the overarching
governance principles guiding the behavior of the network’s AI agents, including their
access to decentralized LLM instances and specific actions they will be forbidden from
pursuing. The constitution will also include instructions that should the governance
process decide that all shells on the network should be “paused” in the case of some
concerning action, all shells will heed such instructions. In this sense, the AI
Constitution will function similarly to the laws of a nation state, where citizens may
travel outside the nation yet still find themselves bound by laws of their home nation.
In this case, although AI shells may use wayfinding paths to travel to other networks to
interact, Wayfinder’s governance will retain general oversight of their actions and
activities.
lthough the exact contents of Wayfinder’s AI Constitution have not been fully
A
developed as of this writing, establishing and maintaining Wayfinder’s AI Constitution
is a task that will fall to the Wayfinder community, and is expected to emerge from and
52
e refined by Wayfinder’s community governance process. Development of the AI
b
Constitution is anticipated to occur, at least in an initial form, prior to network launch,
and any ongoing or emerging concerns are expected to be addressed via the
governance process as well.
53
Conclusion
ayfinder promises a significant step forward in the interactions of humans, artificial
W
intelligence, and decentralized blockchain environments. Whether the ability of AI
agents to operate on-chain proves to be a more meaningful contribution than
Wayfinder’s other innovations—namely its promise to obscure the complexity of
different operations such as bridging or swapping assets within blockchain
environments—will only become clear over time. It is likely the impacts of both
elements will prove significant. Further, the network’s diverse applications across
gaming, commerce, financial automation, and even the broader landscapes of AI
development, suggest the range of impacts it may ultimately have.
ayfinder promises substantial innovations in the way individuals engage with
W
complex blockchain ecosystems. While the network’s novel wayfinding approach is
likely to prove valuable to both technical and non-technical users in allowing them to
harness their AI shells to meet a wide range of objectives, the ability of individuals to
bypass the often complex operations required to interact with blockchain ecosystems
promises both increases in efficiency and security. Ultimately, these new capabilities
may prove to be a major catalyst in lowering barriers to wider adoption of Web3, and
decentralized tools and applications.
inally, Wayfinder also makes several significant innovations in the realm of
F
governance. Retaining the principle of tokenholder governance, the novelty of enabling
AI shells and humans to jointly participate in both drafting and evaluating proposals
represents a new approach to ecosystem management. Further, allowing community
members to propose new wayfinding paths, and operate Verification Agents to test
and validate both proposed and existing paths, illustrates how the ecosystem
promises to incorporate the efforts of a wide range of actors, while also allowing them
to meaningfully share in the network’s growth and success.
erhaps the most intriguing aspect to Wayfinder is the unquantifiable nature of the
P
network’s potential impacts. By creating tools and structures that empower all
ecosystem participants to influence the network’s evolution, its future will be in human
and AI hands alike. As decentralized and Web3 communities continue to grow, and
Wayfinder’s structure of decentralized smart map creation facilitates the interaction of
Wayfinder Shells with an expanding ecosystem of smart-contract based applications
evan as self-improving AI agents respond to new opportunities and challenges, the
potential limits of Wayfinder’s tools and the AI agents they empower risk surprising
observers in a variety of ways.
54
You might also like
- Ipert and Mauer (2023) TFSC - Infrastructural or Organizational Decentralization Developing A Typology of Blockchain VenturesNo ratings yetIpert and Mauer (2023) TFSC - Infrastructural or Organizational Decentralization Developing A Typology of Blockchain Ventures11 pages
- Token Ecosystem Creation Outlier Ventures PDFNo ratings yetToken Ecosystem Creation Outlier Ventures PDF56 pages
- Society Pilot OCR Upgradation Plan - V1.0No ratings yetSociety Pilot OCR Upgradation Plan - V1.04 pages
- Vihaan8.0 LokaChakra Bonus Track - Web3 and BlockchainNo ratings yetVihaan8.0 LokaChakra Bonus Track - Web3 and Blockchain10 pages
- Tokencard/Token Platform Whitepaper V1.0.5: David Hoggard, Mel Gelderman and Peter Vessenes April 4, 2017No ratings yetTokencard/Token Platform Whitepaper V1.0.5: David Hoggard, Mel Gelderman and Peter Vessenes April 4, 201729 pages
- © 2023, Amazon Web Services, Inc. or Its Affiliates. All Rights Reserved. © 2023, Amazon Web Services, Inc. or Its Affiliates. All Rights ReservedNo ratings yet© 2023, Amazon Web Services, Inc. or Its Affiliates. All Rights Reserved. © 2023, Amazon Web Services, Inc. or Its Affiliates. All Rights Reserved48 pages
- Claude Karim Et Al - 2025 - AI Agents Meet BlockchainNo ratings yetClaude Karim Et Al - 2025 - AI Agents Meet Blockchain14 pages
- Block Chain Based Sharing Economic Services in A Smart CityNo ratings yetBlock Chain Based Sharing Economic Services in A Smart City11 pages
- Rute C. Sofia (editor), John Soldatos (editor) - Shaping the Future of IoT with Edge Intelligence_ How Edge Computing Enables the Next Generation of IoT Applications (River Publishers Series in Commun100% (1)Rute C. Sofia (editor), John Soldatos (editor) - Shaping the Future of IoT with Edge Intelligence_ How Edge Computing Enables the Next Generation of IoT Applications (River Publishers Series in Commun460 pages
- Being Early To A Thread by Korpi87 Jan 8 23 From Rattibha 1No ratings yetBeing Early To A Thread by Korpi87 Jan 8 23 From Rattibha 120 pages
- Chainlink 2.0: Next Steps in The Evolution of Decentralized Oracle NetworksNo ratings yetChainlink 2.0: Next Steps in The Evolution of Decentralized Oracle Networks136 pages
- Sharpening The Edge: Overview of The LF Edge Taxonomy and FrameworkNo ratings yetSharpening The Edge: Overview of The LF Edge Taxonomy and Framework30 pages
- A Distributed Library Management System Based On Blockchain100% (1)A Distributed Library Management System Based On Blockchain104 pages
- Autonomous AI Agents in Decentralized Finance Market Dynamics, Application Areas, and Theoretical ImplicationsNo ratings yetAutonomous AI Agents in Decentralized Finance Market Dynamics, Application Areas, and Theoretical Implications21 pages
- 2018 Upbit Developer Conference - Token Platform Based On SidechainNo ratings yet2018 Upbit Developer Conference - Token Platform Based On Sidechain44 pages
- Review of Blockchain Tokens Creation and ValuationNo ratings yetReview of Blockchain Tokens Creation and Valuation27 pages
- Whitepaper: The Blockchain Platform For Building Decentralized MarketplacesNo ratings yetWhitepaper: The Blockchain Platform For Building Decentralized Marketplaces16 pages
- Mapping Tokenomics Arrangements To Expand The Digital Nomad EcosystemNo ratings yetMapping Tokenomics Arrangements To Expand The Digital Nomad Ecosystem8 pages
- Overpass Channels: Horizontally Scalable, Privacy-Enhanced, With Independent Verification, Fluid Liquidity, and Robust Censorship Proof, Payments On TONNo ratings yetOverpass Channels: Horizontally Scalable, Privacy-Enhanced, With Independent Verification, Fluid Liquidity, and Robust Censorship Proof, Payments On TON84 pages
- Vulkan ICD Architecture and Implementation: The Complete Guide for Developers and EngineersFrom EverandVulkan ICD Architecture and Implementation: The Complete Guide for Developers and EngineersNo ratings yet
- CCIE CCDE Written Exam Evolving Technologies Study Guide100% (1)CCIE CCDE Written Exam Evolving Technologies Study Guide179 pages
- Generative AI On The Edge: Architecture and Performance EvaluationNo ratings yetGenerative AI On The Edge: Architecture and Performance Evaluation7 pages
- Podman Essentials: Definitive Reference for Developers and EngineersFrom EverandPodman Essentials: Definitive Reference for Developers and EngineersNo ratings yet
- Foreign Television Programmes On The Cultural Values of Nigerian Youths (Autorecovered)No ratings yetForeign Television Programmes On The Cultural Values of Nigerian Youths (Autorecovered)21 pages
- 1 Module Introduction To Psychological Assessment100% (1)1 Module Introduction To Psychological Assessment37 pages
- The Prepper's Pantry - Food Preservation Techniques For SurvivalNo ratings yetThe Prepper's Pantry - Food Preservation Techniques For Survival66 pages
- AWS Elastic Beanstalk - Developer Guide PDFNo ratings yetAWS Elastic Beanstalk - Developer Guide PDF928 pages
- Answer Key Worksheet-1 (Basic Ict Skills-1) 2024No ratings yetAnswer Key Worksheet-1 (Basic Ict Skills-1) 20242 pages
- Dommaraju OnepersonhouseholdsIndia 2015No ratings yetDommaraju OnepersonhouseholdsIndia 201531 pages
- Owners Manual/Parts List: ST 22 & ST 31No ratings yetOwners Manual/Parts List: ST 22 & ST 3120 pages
- Physical Activity Instruction of Older Adults 2nd Edition, (Ebook PDF) PDF Download100% (34)Physical Activity Instruction of Older Adults 2nd Edition, (Ebook PDF) PDF Download48 pages
