View Project

Long-term memory and retrospection for a local pi.dev agent, backed by Arango. pive-mind copies the agent’s own session logs into a provenance graph, then lets the agent (or you) query its history via AQL and semantic recall — “what commands keep failing?”, “which files do I keep editing?”, “have I solved this before?”.

On top of that provenance it builds an agentic knowledge graph: a structural entity layer mined deterministically at ingest, an optional LLM-backed semantic extractor behind the same contract, and a write-back loop that lets the agent assert facts and merge duplicate entities under a least-privilege curator role — without breaking the replay-safe, read-only-by-construction guarantees.