Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.opencoven.ai/llms.txt

Use this file to discover all available pages before exploring further.

Memory is what turns a harness session into a persistent familiar. OpenCoven treats memory as a first-class concept with four layers:

Working memory

In-session context tied to one PTY run.

Persistent memory

Cross-session memory that survives daemon restarts.

Episodic memory

Remembering specific events, turns, and outcomes.

Semantic memory

Embeddings and concept-based recall.

​
How memory relates to Coven sessions

Coven stores session events — that is the ground truth for what happened. Memory is the layer above that derives, summarizes, and indexes those events for future familiars to consult. The Coven daemon does not enforce a single memory backend. The memory layer is a client concern; Coven only guarantees the event ledger is durable and queryable. Memory search lets a familiar query its own memory inside a session. The exact backend (sqlite-vss, lancedb, a hosted vector store) is up to the client.