greyhaven-ai/autocontext
A recursive self-improving harness that iteratively evaluates AI agents, keeps successful iterations, and can distill learnings into a model for the next agent.
Not currently ranked — collecting fresh signals.
star history
Autocontext is a harness for AI agents that runs iterative evaluation cycles against real tasks. It tracks what worked versus what failed, produces structured traces and artifacts, and optionally distills a local model from successful runs. The system integrates with multiple LLM providers including Claude, Codex, and a local Pi coding agent. Repeated executions improve agent performance over time rather than just varying behavior.
Frequently asked
- What is greyhaven-ai/autocontext?
- A recursive self-improving harness that iteratively evaluates AI agents, keeps successful iterations, and can distill learnings into a model for the next agent.
- Is autocontext open source?
- Yes — greyhaven-ai/autocontext is open source, released under the Apache-2.0 license.
- What language is autocontext written in?
- greyhaven-ai/autocontext is primarily written in Python.
- How popular is autocontext?
- greyhaven-ai/autocontext has 1.3k stars on GitHub.
- Where can I find autocontext?
- greyhaven-ai/autocontext is on GitHub at https://github.com/greyhaven-ai/autocontext.