Open Agents
An open source template for building cloud agents.
Last verified:
What is Open Agents?
Open Agents is an open-source reference application for building and running background coding agents on Vercel's cloud infrastructure. It enables developers to spawn AI coding agents that run infinitely in the cloud, working autonomously on coding tasks without requiring a laptop to stay online. The platform is powered by Vercel's AI SDK, AI Gateway, Sandbox, and Workflow SDK, providing production-grade infrastructure for real agent deployments rather than synthetic demos.
Key features include a full sandbox environment with filesystem, network, and runtime access for each agent; built-in tools for file operations, search, shell commands, and task delegation; Explorer and executor subagents for parallel work; multi-model support through AI Gateway; durable workflows that survive restarts and retry on failure; ephemeral environments with full git integration; auto-hibernation on inactivity with instant restore; and optional auto-commit and pull request creation through GitHub App integration.
Open Agents is designed for developer-tool teams building cloud coding agents into their products, Vercel teams prototyping agent workflows, and engineering leaders evaluating agent architecture. It serves as a forkable starting point that includes the web UI, agent runtime, sandbox orchestration, and GitHub integration needed to go from prompt to code changes. The project uses an MIT license and follows a three-layer architecture separating the web app, durable agent workflow, and sandbox VM.
Open Agents pricing
Pricing model: Freemium
Open Agents is free and open source under MIT license. It operates on a consumption-based model tied to Vercel infrastructure usage where sandbox time and workflow execution are billed based on active compute minutes. The web UI and agent runtime code are free to fork and self-host, but users pay for Vercel Sandbox, Workflow SDK, AI SDK, and AI Gateway usage according to Vercel's pricing. Costs include sandbox provisioning (approximately $0.02 per minute based on observed rates), workflow execution minutes, and AI model token usage through the chosen provider.
Open Agents pros
- Open source with MIT license for commercial use
- Agents run infinitely in cloud without laptop involvement
- Isolated Vercel sandboxes with full filesystem and network access
- Durable workflows that survive restarts and retry on failure
- Built-in file ops, search, shell, and task delegation tools
- Explorer and executor subagents enable parallel work
- Multi-model support with AI Gateway routing and fallbacks
- Auto-hibernation on inactivity with instant restore capability
- Full git integration with automatic commits and pushes
- Optional auto-commit and pull request creation after tasks
- Snapshot and restore filesystem state functionality
- Clear three-layer reference architecture for customization
- Neon Postgres integration for persistence
- Streaming progress updates via Server-Sent Events
- Production-grade primitives from Vercel ecosystem
Open Agents cons
- Vercel-first infrastructure limits portability to other platforms
- Reference app not managed service - users own operations
- GitHub App configuration required for repo access and PRs
- Setup more involved than local-only coding assistants
- Teams outside Vercel must replace key infrastructure pieces
- Operators responsible for database setup and auth settings
- Users must manage sandbox resource profiles and upgrades
- Model-provider policy decisions fall to operators
Frequently asked questions about Open Agents
What is Open Agents?
Open Agents is an open source reference app for building and running background coding agents on Vercel. It includes a web UI, agent runtime, sandbox orchestration, and GitHub integration needed to go from prompt to code changes without keeping your laptop involved.
Does Open Agents run the agent inside the sandbox?
No. The agent runs outside the sandbox and interacts with it through tools such as file reads, edits, search, and shell commands. The sandbox stays focused on providing filesystem, shell, git, dev servers, and preview ports.
Can Open Agents create pull requests?
Yes, Open Agents supports optional auto-commit, push, and PR creation after a successful run. Those capabilities depend on the GitHub App setup and user preferences.
What license does Open Agents use?
Open Agents is licensed under the MIT License, making it a permissive starting point for teams that want to fork, modify, and adapt the reference app for commercial projects.
How do you run Open Agents locally?
The README documents a Bun-based local setup: run 'bun install' to install dependencies, copy apps/web/.env.example to apps/web/.env, fill in required environment values, then run the web app with 'bun run web'.
Who is Open Agents for?
Open Agents fits developer-tool teams adding background coding agents to a developer product, Vercel teams prototyping agent workflows, and engineering leaders evaluating agent architecture. It is not for users wanting a finished managed coding assistant rather than a reference app to adapt.
What infrastructure does Open Agents use?
Open Agents is built on Vercel's production-grade primitives including AI SDK for unified model access, AI Gateway for routing with fallbacks and rate limiting, Sandbox for secure isolated environments, and Workflow SDK for durable resumable agent workflows.
How does the agent architecture work?
The three-layer architecture separates the Next.js web app (managing chat, auth, sessions), the agent workflow engine using Vercel Workflows for durable background execution, and isolated VM sandboxes for code execution with filesystem, shell, and dev server access.
What tools can agents use in Open Agents?
Agents have access to file read/write operations, grep and glob search, bash shell commands, and task delegation. The Tools Library provides JSON schemas for these operations, and the LLM orchestrates tool calls through a ToolLoopAgent.
Can I deploy Open Agents to Vercel?
Yes, the project ships with a Vercel deploy flow including Neon Postgres integration support and explicit environment variable setup steps, giving teams a path from fork to hosted reference app.