cc-hybrid-multiagent
A self-improving multi-agent framework: pinned judge, cross-check on a second model family, regression gate that refuses a single-repeat run, and an autonomy ladder that starts at HOLD.
Everything here is something I built and ran, not a tutorial project. The posts on this site are mostly about how these broke.
A self-improving multi-agent framework: pinned judge, cross-check on a second model family, regression gate that refuses a single-repeat run, and an autonomy ladder that starts at HOLD.
A fully local agent stack on Apple Silicon. Container-isolated harness, in-process PII masking before any trace leaves the box, and an intake gate where review is the default state rather than a phase.