Provena

Provena is an open-source, evidence-backed memory layer for AI agents. It records immutable events, turns durable facts into scoped claims, and preserves the evidence behind every memory. Codex, Claude, Gemini, and generic MCP clients can share context across sessions while users can inspect why a claim exists, review candidate memories, trace conflicts, and reverse decisions. Provena runs locally with PostgreSQL and supports local Ollama models, keeping memory auditable and under your control.