Brevet Brings Change Control to Agent Learning
Brevet is an open-source, local-first Python runtime that governs how AI agents evolve without changing model weights. It turns corrections into candidate capabilities, requires human promotion, and ships learned behavior in signed, versioned releases that can be evaluated, audited, or recalled.
Brevet targets the missing control plane between agent memory and production software governance. Its strongest idea is treating learned behavior as deployable code rather than invisible prompt drift.
- –Records corrections, provenance, approvals, and evidence in a hash-linked chain
- –Converts recurring overrides into proposals without granting them immediate authority
- –Replays evaluations before promotion to catch regressions
- –Supports rollback and recall when a learned capability proves unsafe or ineffective
- –Wraps existing Python agent frameworks, including LangGraph, CrewAI, and the OpenAI Agents SDK
DISCOVERED
1h ago
2026-10-14
PUBLISHED
1h ago
2026-10-14
RELEVANCE
AUTHOR
ShinkaIoT