Episodes

Episodes are private immutable, scrubbed local records for bounded recall. They capture a useful summary with safe provenance; they are not a silent copy of chat history or terminal output.

Retrieval boundaries

ebrain episodes list --json
ebrain episodes recall <query> --json
ebrain episodes get <episode-id> --json

List and recall are summary/bounded surfaces. An episode body needs an explicit get and remains character-bounded. Public JSON contracts reject filesystem paths, raw bodies on passive surfaces, and unrelated private metadata.

Recording and mirroring

An approved ebrain remember learning can mirror to bounded local episode recall on a best-effort basis after its durable learning write. If mirroring fails, the durable learning stays valid; a secondary recall feature never rolls back a successfully written decision.

Explicit episode record actions validate generated workspace identity, size, safe provenance, private permissions, and secret/denied-source boundaries. Episode files are immutable in practice: a record creates a new opaque identifier rather than editing old content.

What episodes do not claim

• They do not preserve every message from a session.
• They do not automatically train or change an agent.
• They do not write to a federated source without its existing approved path.
• They do not provide a public import of personal historical data.

The only legacy recovery proof is fixture-only and synthetic. See migration concepts.