Hi Product Hunt 👋 I started building Provena after seeing how quickly agent memory can become a black box. A retrieved memory may be relevant, but most systems cannot clearly answer who asserted it, when it was valid, what evidence supports it, or what contradicts it. Provena models memory as claims backed by immutable evidence. Events are preserved, claims are scoped and versioned, and candidate memories can be reviewed instead of silently becoming trusted facts. It integrates through MCP with Codex, Claude, Gemini, and custom agents. It can also use local Ollama models for fact extraction and embeddings, with PostgreSQL as the system of record. Provena is currently in alpha, and I would appreciate practical feedback on installation, automatic conversation capture, retrieval quality, and the claim-explain workflow. The project is open source, and we have several clearly scoped good-first issues for contributors. GitHub: https://github.com/admiralpunk/P... PyPI: https://pypi.org/project/provena... Thank you for taking a look.

Provena

Evidence-backed memory for AI agents

Evidence-backed memory for AI agents