E2B

Update: Domain-Based Firewall ☄️ Secure your sandbox environment by restricting internet access to specific hostnames or wildcard patterns. This update gives you granular control over outbound traffic, allowing you to easily define trusted domains and ensure your workloads only communicate with authorized external services.

Excited to share that ZeroTwo was accepted into the E2B for Startups program. We received $20K in E2B credits + E2B Pro Tier. This is a major stepping stone in expanding the capabilities for our agents, and providing an expanded free tier to our users! Big thanks to the team at E2B, can't wait to continue exploring use cases of their sandbox environment made for AI agents! If you are an early startup, check out if you are eligible for the program! --> https://e2b.dev/startups

The E2B for Startups applications for first 2026 batch are open! The program offers early-stage companies and teams $20,000 in E2B credits and full access to the Pro Tier to build, test, and run AI agents in secure cloud sandboxes. E2B is trusted by many leading AI-first companies like Manus, Rork, Perplexity or Hugging Face, as well as YC companies such as Delve or Gumloop. The program is designed to help teams grow and scale on an open-source, AI-first infrastructure. To apply, sign in to E2B and use your Team Id in the application. If you want priority approval, comment or repost this. We will review these first and will try to get to everyone fast. Happy building! 🫡

The E2B for Startups applications for first 2026 batch are open! The program offers early-stage companies and teams $20,000 in E2B credits and full access to the Pro Tier to build, test, and run AI agents in secure cloud sandboxes. E2B is trusted by many leading AI-first companies like Manus, Rork, Perplexity or Hugging Face, as well as YC companies such as Delve or Gumloop. The program is designed to help teams grow and scale on an open-source, AI-first infrastructure. To apply, sign in to E2B and use your Team Id in the application. If you want priority approval, comment or repost this. We will review these first and will try to get to everyone fast. Happy building! 🫡

Back in 2023, when Tomas Valenta and I started E2B, we had one core hypothesis: Coding agents are the general agents because code is a general tool. Therefore, if we build the best runtime for coding agents, we build the best runtime for all AI agents and sandbox is the core building block. That hypothesis will become a reality this year. Coding agents are completely versatile - generating slides, videos, ads, and of course building close to any software you want. If 2025 was a ride, 2026 will be a spacelift.

New year, new batch of builders. 🎇 We are re-opening applications for E2B for Startups. We have supported hundreds of early teams, companies like Rork or YC startups like Rowboat (YC S24), Chima or Twenty. Program members get free credits and extra benefits to build the best product with the E2B AI cloud. If you want priority approval, comment or repost this. I’ll review these first and will try to get to everyone fast. 💨 Let's make 2026 the year of building!

Behind the scenes of Vibe Coding Agents: Code Sandboxes explained (E2B as reference) Recently, I started digging into the infrastructure behind Vibe Coding Agents and wanted to share what I learned about Code Sandboxes. Big realization: Docker and Lambda weren’t designed for how AI thinks. AI agents need safe, stateful, pauseable environments — not stateless containers. Here’s what’s actually happening under the hood 👇 (and why users feel the difference) 1️⃣ Firecracker > Docker Instead of containers, these platforms use microVMs (Firecracker). • Hardware-level isolation → safe for untrusted AI-generated code • Sub-200ms startup via pre-warmed VM pools 👉 This is why AI coding tools feel responsive, not “run → wait → reload”. You type → code runs → preview updates instantly. 2️⃣ Serverless with RAM (Snapshotting) They don’t just stop code — they snapshot the entire running machine (CPU + RAM). • Python objects, variables, dev servers stay in memory • Resume hours later exactly where you left off 👉 You can leave a vibe-coding or data analysis session, come back later, and everything is still alive. No reloading data. No re-running setup. 3️⃣ Copy-on-Write Filesystems No Docker image pull per run. • Base VM images already exist on the host • New sandboxes only store what changes 👉 Heavy environments (10GB+ Python/Node stacks) start in milliseconds. No “installing dependencies…” anxiety. Why this category exists (real use cases) This infrastructure unlocks things that were previously painful or impossible: • Vibe Coding IDEs → Cursor / Replit-style agents that write + run code live • AI Data Analysts → Keep large datasets in memory across chat turns • LLM Evals & Testing → Safely run thousands of generated programs in parallel • Agent Platforms → Where an agent owns a long-lived computer, not just tools Market landscape (rough buckets) • E2B → Developer-first agent sandboxes • Modal → Serverless compute + sandboxes for AI workloads • Replit → Vertically integrated (IDE + runtime) • Docker / Lambda → Excellent infra, wrong abstraction for agents Special thanks to Tomáš Beran and Tereza Tizkova for supporting this post

Vibe3 sandboxes now run on E2B Every onchain app you create runs in its own secure, private cloud workspace. What this means for builders: ✔️ Go from idea to app up to 50× faster ✔️ Your app runs in a safe, private environment ✔️ Watch AI build your onchain app in real time Start building → vibe3.build

Congratulations & proud to be supporting Manus team with E2B virtual computers since earlier days. 🤝