Episode Workflows

Episodes let you group related memories into narrative sessions. When you’re debugging a tricky issue, implementing a feature, or doing a code review, episode memory captures the full story — not just individual facts.

When to Use Episodes

• Debugging sessions — Chasing a bug across multiple files and hypotheses
• Feature implementations — Multi-step work where context builds over many memories
• Code reviews — Reviewing a PR with findings that build on each other
• Research spikes — Exploring a new technology or approach

Starting an Episode

{
  "tool": "begin_episode",
  "title": "Debugging auth token expiry",
  "project": "my-project"
}

All memories stored after begin_episode are automatically grouped into the episode.

Ending an Episode

When the work session wraps up, end the episode with a summary:

{
  "tool": "end_episode",
  "summary": "Root cause was refresh token rotation race condition. Fixed by adding mutex around token refresh. Also discovered the token cache wasn't clearing on logout."
}

The summary is the main value for future recall — it captures what was attempted, what worked, and what was decided.

Recalling Past Episodes

When you need to revisit a past session:

{
  "tool": "list_episodes",
  "project": "my-project"
}

Then retrieve a specific episode:

{
  "tool": "recall_episode",
  "id": 5
}

This returns the episode’s memories in order, plus the summary.

Best Practices

• Give episodes descriptive titles (“Debugging auth token expiry”, not “Session 1”)
• Always provide a summary when ending — this is what future recall surfaces
• Episodes auto-end after 30 minutes of inactivity
• Don’t start episodes for routine Q&A or single-memory interactions