What you're trying to do
gbrain's docs/guides/push-context.md documents three push channels — ambient reflex, volunteer_context op, gbrain watch — through which the brain volunteers confidence-gated pages into the conversation window. Today vouch is purely pull: the agent asks, vouch answers. Adding a push channel lets vouch surface "you have an approved claim about this; consider citing it" mid-session.
Suggested shape
- New
kb.volunteer_context MCP op (server → client streaming notification).
- Server-side trigger:
kb.session_start opens a watch on hot-memory + salience deltas; when a claim's relevance score for the active session exceeds a config threshold (default 0.85), push a notification with {claim_id, relevance, why}.
- Throttled to one push per 30 seconds; rate-limited per session.
Acceptance
Out of scope
- Two-way conversational volunteering — push only.
- Volunteering proposals (not just durable claims).
What you're trying to do
gbrain's
docs/guides/push-context.mddocuments three push channels — ambient reflex,volunteer_contextop,gbrain watch— through which the brain volunteers confidence-gated pages into the conversation window. Today vouch is purely pull: the agent asks, vouch answers. Adding a push channel lets vouch surface "you have an approved claim about this; consider citing it" mid-session.Suggested shape
kb.volunteer_contextMCP op (server → client streaming notification).kb.session_startopens a watch on hot-memory + salience deltas; when a claim's relevance score for the active session exceeds a config threshold (default 0.85), push a notification with{claim_id, relevance, why}.Acceptance
Out of scope