signet

# Signet Signet is the local substrate for agent continuity. It is not just a memory app, and it is not a hidden daemon that gets to rewrite truth on its own. The practical boundary is: - sources and transcripts remain inspectable evidence - memories are scoped, searchable recall rows - ontology stores reviewed structure, currentness, versions, and links - epistemic assertions preserve who claimed or believed something - skills own repeated procedures - identity and AGENTS files hold operating policy - secrets stay out of chat, memory, logs, and source files - connectors keep harnesses attached to the same local substrate When Signet behavior is unclear, inspect the real checkout, daemon, API, database-derived outputs, and installed harness config before guessing. ## Current Mental Model Use this model when explaining or operating Signet: 1. **Source truth**: transcripts, source artifacts, imported files, notes, configs, and documents are evidence. They should keep provenance, timestamps, paths, and ids. 2. **Recall**: `POST /api/memory/recall` is the canonical explicit recall surface. It preserves score, source, type, supplementary context, and session-dedupe metadata. 3. **Remember**: `POST /api/memory/remember` writes explicit scoped memory rows. Raw remember does not create arbitrary ontology structure. 4. **Ontology**: graph and knowledge updates go through audited `signet ontology ...` operations with evidence, actor, confidence, and version/...