The Raw Stream
Episodes are the raw stream of everything that happened. Every conversation turn gets a one-line summary appended to today’s episode file. No filtering, no summarization, no LLM call — just a formatted timestamp and description.
Location
~/.wolffish/workspace/brain/hippocampus/episodes/
├── 2026-05-14.md
├── 2026-05-15.md
└── 2026-05-16.md
YYYY-MM-DD.md.
How Episodes Are Written
After every turn completes, hippocampus.appendEpisode() writes a single line to today’s file:
// Simplified from src/main/runtime/hippocampus.ts
async appendEpisode(summary: string): Promise<void> {
const today = format(new Date(), 'yyyy-MM-dd')
const file = join(this.episodesDir, `${today}.md`)
const line = `- ${format(new Date(), 'HH:mm')} — ${summary}\n`
await appendFile(file, line)
this.corpus.emit('memory.episode.appended', { file, summary })
}
No LLM call is involved. The summary is derived from the conversation turn directly — it’s a deterministic extraction of what happened, not a generated summary. This keeps episode logging instant and free.
Example Episode File
Here’s what 2026-05-16.md might look like after a day of use:
# Episodes — 2026-05-16
- 08:32 — Asked about React 19 Server Components, discussed streaming patterns
- 08:45 — Generated a useServerAction hook with error boundaries
- 09:10 — Reviewed PR #247, suggested simplifying the auth middleware
- 11:30 — Searched for flights to Riyadh, compared prices on Google Flights
- 14:15 — Explained the difference between consolidation and knowledge promotion
- 16:02 — Created a new capability plugin for Notion integration
- 17:40 — User said "remember I prefer pnpm over npm" → updated preferences.md
How Episodes Are Used
Context Assembly
The prefrontal module loads the most recent 3 days of episodes by default during context assembly. These pass through the RAS for relevance scoring — only entries that match the current conversation topic make it into the final context.
Episodes are the raw input for nightly consolidation. The brainstem scheduler reads episode files, sends them to an LLM for summarization, and produces a compressed weekly file. After consolidation, episodes remain on disk (they’re never deleted automatically).
Cortex Search
All episode files are indexed by the cortex (SQLite FTS5). When you ask “what did we discuss about auth last week?”, the cortex searches episode entries and returns matching fragments with BM25 relevance ranking.
Manual Editing
Episode files are plain markdown. You can:
- Delete entries you don’t want remembered
- Add entries to seed context (e.g., ”- 09:00 — Started the new billing service project”)
- Edit entries to correct inaccuracies
Changes are picked up by the brainstem file watcher and re-indexed in the cortex automatically.
If you delete an episode file entirely, that day’s history is gone from memory. The cortex will remove it from the search index on the next file-watch cycle.
Configuration
Episodes are always written — there’s no toggle to disable them. They’re the foundation of the entire memory system. However, you can control how many days of episodes load into context:
// config.json
{
"memory": {
"episodeDaysInContext": 3
}
}
Increase episodeDaysInContext if you have long-running projects and want more historical context. Decrease it if your conversations are mostly independent and you want to save token budget for other memory tiers.
