Agenta-AI/agenta

What it solves

Agenta provides a centralized workspace for building, running, and managing specialized AI agents that automate work. It bridges the gap between simple chat interfaces and rigid workflow builders by allowing agents to plan, use tools, and adapt their approach while providing the necessary infrastructure for team collaboration, versioning, and background execution.

How it works

Users build agents through a conversational interface, describing the desired work and connecting the necessary applications. The platform supports multiple "harnesses" (like Claude Code, Pi, and Codex) and a wide array of LLM providers (OpenAI, Anthropic, Ollama, etc.). Agents can be deployed as interactive chat assistants or as background agents that trigger based on schedules or events. The system utilizes MCP servers and skills to extend agent capabilities and integrates with over 1,000 apps via Composio.

Who it’s for

It is designed for individuals and teams who want to create autonomous agents that can handle non-predictable tasks requiring judgment, as well as those who want to self-host their agent infrastructure to utilize existing LLM subscriptions without relying solely on metered API billing.

Highlights

• Hybrid Execution: Run agents interactively in chat or automatically in the background via schedules and event triggers.
• Model & Harness Flexibility: Switch between different LLM providers and execution harnesses without rebuilding the agent.
• Open Standards: Uses AGENTS.md, skills, and MCP servers for agent definitions.
• Observability: Full tracing of model and tool calls, version history for configurations, and cost/token tracking.
• Collaborative Workspaces: Shared file environments where humans and agents can co-author documents or research.
• Human-in-the-loop: Granular permission settings to determine which tool actions are automatic and which require human approval.