The open-source workspace for your agents.

Build agents through chat. Improve them with feedback. Share them with your whole team.

Engineering

Code review agent

Reviews every pull request and leaves inline comments.

Trigger
On every pull request

Harness
Claude Code
Codex
pi.dev

Models
Self-hosted

Skills
review-diff
style-check
risk-scan

Tools
GitHub
Linear

AGENTS.md

# Code review agent

You review pull requests for this repository. Be concise and specific. Never approve your own suggestions.

## When triggered  
A pull request is opened or a new commit is pushed to an open PR.

## Steps  
1. Read the PR title, description and linked Linear issue  
2. Review the full diff, file by file  
3. Summarize what changed and why, in 3 sentences max  
4. Flag bugs, security issues and perf regressions inline  
5. Suggest concrete, minimal fixes with code snippets

## Rules  
- Comment only where you would block a merge  
- Prefer one high-signal comment over five nitpicks  
- Never rewrite whole files, propose diffs  
- Style issues: link the guide, do not lecture

## Escalation  
Tag the code owner when the diff touches auth, billing or migrations.

Open standards

One agent package, any stack

Your agent is AGENTS.md, skills, and MCP tools. Swap the harness, model, or runtime without touching a line. When self-hosting locally, you can use your Claude or ChatGPT subscription.

Skills
review-diff
classify
research-company
pull-metrics
style-check
answer-cited
draft-outreach
risk-scan
build-dashboard

Human in the loop

Consequential actions wait for you. Approve, edit, or reject before anything ships.

Version every change

Prompts, skills, and tools are versioned like code. Roll back to any point, anytime.

Trace every run

See each step, its cost, and its outcome — and find the exact failure points.

Improve continuously

Feedback becomes test cases; every change is evaluated before it ships.

Open source

Self-host Agenta on your infrastructure

MIT-licensed and yours to run. Keep your agents and data on your own premises — same code as Agenta Cloud, no lock-in.

• MIT license
• Docker Compose
• Helm charts

Self-host — with your coding agent

Copy prompt

paste into Claude Code — it walks you through setup and testing

$ npx skills add Agenta-AI/agenta-skills > Help me
self-host Agenta with its repository. → Agenta running at agenta.acme.com

Ship agents that actually work

Build with skills and tools, run on any harness in any environment, and improve with real feedback. All open source.