Agenta-AI/agenta

The open-source workspace for building and running AI agents. Build agents through chat, share them with your team, and run background agents on schedules or app events.

What it solves

Agenta provides a centralized workspace for building, running, and managing AI agents that automate work. It addresses the difficulty of moving from a simple chat interface to reliable, background-running agents that can interact with external applications and be shared across a team.

How it works

Users create agents by chatting with them, describing the required work, and connecting necessary applications. Agenta supports a variety of model providers (e.g., OpenAI, Anthropic, Ollama) and "harnesses" (like Claude Code and Pi) to execute the agent's logic. It uses open standards like AGENTS.md, skills, and MCP servers to define agent capabilities and integrate with over 1,000 apps via Composio.

Who it’s for

Developers and teams who want to build specialized AI agents that can run either interactively in a chat or automatically in the background on a schedule or event-trigger, without being locked into a single model or platform.

Highlights

  • Flexible Model Support: Supports a wide range of API-based and self-hosted models.
  • Background Execution: Ability to run agents on a schedule or when specific events occur in connected apps.
  • Harness Switching: Change the execution layer (harness) without rebuilding the agent.
  • Observability: Full tracing of model and tool calls, including token usage and cost tracking.
  • Human-in-the-loop: Granular permissions for tool use, allowing for automatic execution or required human approval.
  • Shared Workspaces: Collaborative environments where agents and humans can work together on files and research.