How Tavily Uses MongoDB to Enhance Agentic Workflows
As AI agents grow in popularity and are used in increasingly mission-critical ways, preventing hallucinations and giving agents up-to-date context is more important than ever. Context can come from many sources—prompts, documents, proprietary internal databases, and the internet itself. Among these sources, the internet stands out as uniquely valuable, a best-in-class resource for humans and LLMs alike due to its massive scale and constant updates. But how can
large language models
(LLMs) access the latest and greatest information from the internet?
Enter
Tavily
, one of the companies at the heart of this effort. Tavily provides an easy way to connect the web to LLMs, giving them the answers and context they need to be even more useful. MongoDB had the opportunity to sit down with Rotem Weiss, CEO of Tavily, and Eyal Ben Barouch, Tavily’s Head of Data and AI, to talk about the company’s history, how Tavily uses MongoDB, and the future of agentic workflows.
Tavily’s origins
Tavily began in 2023 with a simple but powerful idea. "We started with an open source project called
GPT Researcher
," Weiss said. "It did something pretty simple—go to the web, do some research, get content, and write a report." That simplicity struck a chord. The project exploded, getting over 20,000 GitHub stars in under two years, signaling to the team that they had tapped into something developers desperately needed.
The viral success revealed a fundamental gap in how AI systems access information. "So many use cases today require real-time search, whether it's from the web or from your users," Weiss noted. "And that is basically
RAG (retrieval-augmented generation)
."
"Developers are slowly realizing not everything is semantic, and that vector search alone cannot be the only solution for RAG," Weiss said.
Indeed, for certain use cases, vector stores benefit from further context. This insight, buttressed by breakthrough research around
CRAG (Corrective RAG)
, pointed toward a future where systems automatically turn to the web to search when they lack sufficient information.
Solving the real-time knowledge problem
Consider the gap between static training data and our dynamic reality. Questions like "What is the weather today?" or "What was the score of the game last night?" require an injection of real-time information to accurately answer. Tavily's system fills this gap by providing AI agents with fresh, accurate data from the web, exactly when they need it.
The challenge Tavily addresses goes beyond information retrieval. “Even if your model ‘knows’ the answer, it still needs to be sent in the right direction with grounded results—using Tavily makes your answers more robust,” Weiss explained.
The new internet graph
Weiss envisions a fundamental shift in how we think about the architecture of the web. "If you think about the new internet, it’s a fundamentally different thing. The internet used to be between people—you would send emails, you would search websites, etc. Now we have new players, the AI agents, who act as new nodes on the internet graph."
These new nodes change everything. As they improve, AI agents can perform many of the same actions as humans, but with different needs and expectations. "Agents want different things than people want," Weiss explained. "They want answers; they don't need fancy UIs and a regular browser experience. They need a quick, scalable system to give them answers in real time. That's what Tavily gives you."
The company's focus remains deliberately narrow and deep. "We always want to stick to the infrastructure layer compared to our competitors, since you don't know where the industry is going," Weiss said. "If we focus on optimizing the latency, the accuracy, the scalability, that's what is going to win, and that's what we're focused on."
Figure 1.
The road to insightful responses for users with TavilyHybridClient.
MongoDB: The foundation for speed and scale
To build their infrastructure, Tavily needed a database that could meet their ambitious performance requirements. For Weiss, the choice was both practical and personal.
"MongoDB is the first database I ever used as a professional in my previous company," he said. "That's how I started, and I fell in love with MongoDB. It's amazing how flexible it is–it's so easy to implement everything." The document model, the foundation upon which MongoDB is built, allowed Tavily to build and scale an enterprise-grade solution quickly.
But familiarity alone didn't drive the decision.
MongoDB Atlas
had the performance characteristics Tavily required. "Latency is one of the things that we always optimize for, and MongoDB delivers excellent price performance," Tavily’s Ben Barouch explained. "The performance is much more similar to a hot cache than a cold cache. It's almost like it's in memory!"
The managed service aspect proved equally crucial. "MongoDB Atlas also saves a lot of engineering time," Weiss noted. In a fast-moving startup environment, MongoDB Atlas enabled Weiss to focus on building Tavily and not worry about the underlying data infrastructure. "Today, companies need to move extremely fast, and at very lean startups, you need to only focus on what you are building. MongoDB allows Tavily to focus on what matters most, our customers and our business."
Three pillars of success
The Tavily team highlighted three specific MongoDB Atlas characteristics that have become essential to their operations:
Vector search
: Perhaps most importantly for the AI era, MongoDB's vector search capabilities allow it to be "the memory for agents." As Weiss put it, "The only place where a company can have an edge is their proprietary data. Every company can access the best models, every company can search the web, every company can have good agent orchestration. The only differentiation is utilizing your internal, proprietary data and injecting it in the fastest and most efficient way to the prompt."
MongoDB, first with
Atlas Vector Search
and now with
Hybrid Search
, has effective ways of giving agents performant context, setting them apart from those built with other technologies.
Autoscaling
: "Our system is built for a very fast-moving company, and we need to scale in a second," Weiss continued. "We don't need to waste time each week making changes that are done automatically by MongoDB Atlas."
Monitoring
: "We have other systems where we need to do our own monitoring with other cloud providers, and it's a lot of work that MongoDB Atlas takes care of for us," Weiss explained. "MongoDB has great visibility."
Betting on proven innovation
Tavily has been impressed with the way MongoDB has kept a finger on the pulse of the evolving AI landscape and added features accordingly. “I believed that MongoDB would be up to date quickly, and I was right," Weiss said. "MongoDB quickly thought about vector search, about other features that I needed, and got them in the product. Not having to bolt-on a separate vector database and having those capabilities natively in Atlas is a game changer for us."
Ben Barouch emphasized the strategic value of MongoDB’s entire ecosystem, including the community built around the database: "When everyone's offering the same solutions, they become the baseline, and then the things that MongoDB excels at, things like reliability and scalability, are really amplified. The community, especially, is great; MongoDB has excellent developer relations, so learning and using MongoDB is very easy."
The partnership between MongoDB and Tavily extends beyond technology to trust. "In this crazy market, where you have new tools every two hours and things are constantly changing, you want to make sure that you're choosing companies you trust to handle things correctly and fast," Weiss said. "I want a vendor where if I have feedback, I'm not afraid to say it, and they will listen."
Looking ahead: The multi-agent future
As Tavily continues building the infrastructure for AI agents to search the web, Weiss sees the next evolution already taking shape. "The future is going to be thinking about combining these one, two, three, four agents into a workflow that makes sense for specific use cases and specific companies. That will be the new developer experience."
This vision of orchestrated AI workflows represents just the beginning. With MongoDB Atlas providing the scalable, reliable foundation they need, Tavily is positioning itself at the center of a fundamental shift in how information flows through our digital world.
The internet welcomed people first, then connected them in revolutionary ways. Now, as AI agents join the network, companies like Tavily are building the infrastructure to ensure this next chapter of digital evolution is both powerful and accessible. With MongoDB as their foundation, they're not just adapting to the future—they're building it.
Interested in building with MongoDB Atlas yourself?
Try it today
!
Use Tavily for working memory in this
MongoDB tutorial
.
Explore Tavily’s
Crawl to RAG
example.
August 5, 2025
