overmind
View on GitHubThe platform for continuously improving AI agents.
A self-hostable platform for improving AI agents from production traces: instrument and evaluate agents, build datasets, fine-tune and serve models, and test prompt or code changes. Includes an MCP server for agent workflows.
Use Cases
Instrument agents and collect production tracesEvaluate agent capabilities on live traces or datasetsCreate versioned training datasets from tracesFine-tune and deploy models for specialized agentsExperiment with prompts, tools, and agent control flowConnect coding agents to platform workflows through MCP
Built With
- Language
- Python
- Frameworks
- Django · OpenTelemetry · Celery · MCP · Modal · vLLM
Tags
AI agents · agent training · LLMOps · evaluations · fine-tuning · observability · production traces · MCP · datasets · model serving