Vibeleaderboard
Index / tool
Visit open-agents.dev
Category
AI Agents
Rank
Pricing
Open Source
Type
TOOL
Builder
@rauchg
Date

About

An open-source framework for building background coding agents that can make code changes to GitHub repositories autonomously. Provides a web UI, agent runtime, and sandbox orchestration for creating agents that work independently of your laptop.

What it does

Open Agents turns a chat request into a durable workflow that directs a separate cloud sandbox. The workflow reads and edits files, searches code, runs shell commands, and can delegate tasks. The sandbox holds the repository, branch, development servers, and preview ports. Runs stream progress to the web interface, survive reconnection, and can optionally finish by committing, pushing, and opening a pull request.

Why it's ranked here

The architecture makes sense for work that outlives a browser request. Durable workflow steps, reconnectable streams, cancellation, sandbox hibernation, and snapshot restoration address real failure and continuity problems. The repository also exposes agent and sandbox packages instead of hiding everything behind the application. The verdict is positive for teams prepared to own a Vercel-centered reference implementation, but less compelling for anyone seeking a ready-managed service or broad sandbox portability.

What's good

The control logic stays outside the virtual machine, so agent execution and sandbox lifetime can change independently. Git handling preserves local edits while synchronizing remote branches, validates branch names, recognizes binary files, and retains executable modes. Model settings support provider-specific defaults and overrides while forcing OpenAI response storage off. Read-only session sharing, optional pull request creation, cancellation, preview ports, skills, subagents, and usage accounting make the reference app unusually complete.

Tradeoffs

Deployment requires Postgres and session secrets, then separate Vercel OAuth and GitHub App configuration for the full repository workflow. Voice input adds another provider credential. The sandbox abstraction currently resolves only to Vercel, so portability is architectural intent rather than demonstrated support. The project expects consumers to fork and adapt it, which means they own deployment, credentials, policies, and product changes. Automatic commits and pull requests are optional preferences, not a guaranteed completion path.

How to use it well

Use it when a team wants to own a chat-based coding service whose jobs continue after users disconnect. Start from the reference application, configure authentication and repository access, then adapt agent instructions, skills, models, resource profiles, and sandbox hooks to your workflow. Enable automatic commits or pull requests only where that policy fits. It does not replace a fully managed coding-agent product, a provider-neutral sandbox fleet, or a simple local assistant for short interactive edits.

Technical notes+

The root package.json defines a private ESM pnpm 11 monorepo on Node 24, with Turbo builds and Bun-backed isolated tests. packages/agent/open-agent.ts constructs openAgent as an AI SDK ToolLoopAgent, injects sandbox context through validated call options, and registers file, shell, search, task, skill, question, and web-fetch tools. packages/agent/models.ts wraps gateway models with provider defaults and enforces store: false for OpenAI. packages/sandbox/factory.ts dispatches only Vercel state to connectVercel; packages/sandbox/interface.ts defines filesystem, execution, lifecycle, timeout, port, snapshot, and state contracts. packages/sandbox/git.ts uses NUL-delimited Git output, temporary credential brokering, and stash-reset-pop recovery. Tests are colocated in packages/agent/models.test.ts and packages/sandbox/git.test.ts.

Observed

Primary language
TypeScript
Packaging
Private ESM pnpm monorepo with Turbo orchestration
Runtime
Node.js 24.x
Interfaces
Next.js web application plus exported agent and sandbox TypeScript packages
Sandbox platform
Vercel Sandbox is the only implemented sandbox state discriminator
Installation surface
Forkable Vercel deployment or local pnpm installation
Testing
Bun tests are colocated with agent and sandbox source files

Read from README.md, package.json, packages/agent/index.ts, packages/agent/types.ts, packages/agent/usage.ts, packages/sandbox/git.ts, packages/shared/index.ts, packages/agent/models.ts, packages/sandbox/index.ts, packages/sandbox/types.ts, packages/sandbox/factory.ts, packages/agent/open-agent.ts, packages/sandbox/git.test.ts, packages/agent/models.test.ts, packages/sandbox/interface.ts.

What it can do

  • Create autonomous coding agents for GitHub repositories

    Agent configuration and GitHub repository accessBackground coding agent that can operate independently

  • Execute code changes in isolated sandbox environments

    Code modification requests and target repositoryCode changes applied to GitHub repository

  • Orchestrate multiple sandbox virtual machines

    Agent workflow requirements and resource specificationsManaged sandbox instances for code execution

  • Manage agent workflows through web interface

    User commands and agent parameters via web UIConfigured and monitored agent operations

  • Process GitHub repository operations autonomously

    Repository URL and coding task specificationsAutomated commits, pull requests, or code modifications

Tags

agentsaiautomationautonomouscloudcodingcoding-agentgithubopen-sourcesandboxvercelworkflow

Tech Stack

Node.jsTypeScript

Media

Open Agents

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.