Agenta

Agenta is an open-source workspace where you build AI agents through chat, refine them using feedback, and share them across your team. It’s designed for teams that want agents to run in their own environment (self-host) or through a cloud setup, while keeping agent behavior and actions under human control.

It’s a good fit if you’re responsible for agent quality, day-to-day reliability, and internal reuse. You can start from templates, then tailor agents to your workflows and data, with a consistent package format (AGENTS.md, skills, and tools).

• Chat-based agent building: draft and adjust agents directly in the workspace, using your context and integrations
• Templates for common roles: engineering (code review), support, sales, knowledge, and operations starting points you can customize
• Scheduled automation: tell an agent when to run and let it keep doing the job, while still supporting review for consequential actions
• Human-in-the-loop approvals: certain actions pause for your approval, so edits and rejects happen before anything ships
• Versioning and rollback: prompts, skills, and tools are versioned like code, so you can revert to a known state
• Traceable runs: see each step outcome so failures and costs are easier to pinpoint
• Feedback as test cases: improvements can be evaluated before changes go out