Briefing — an assembled answer

A briefing is a single organized response that pulls together everything relevant to your question. Not a list of search results — a one-shot summary of how your company currently knows the topic.

What you get back

For a question like "what's our rate limit policy for the payments service?", the briefing pulls together:

Current policy — what's in effect right now
Recent changes — what shifted in the last few days/weeks
Conflicts — when different sources disagree, all sides shown
Owners — who manages this area
Gaps — places where there isn't enough material to be confident

Not five document hits — a one-page picture of the company's current position.

Who uses it

1. People. Before making a decision, quickly check "what are we doing right now?"

2. AI agents. Tools like Claude or Cursor call this automatically before doing work, so they don't answer with generic content that ignores your company context.

Two modes

Mode	When to use	Speed / cost
Quick	Day-to-day work, fast context check	Instant, free
Deep	High-stakes decisions, writing, external responses	1-2 seconds, small cost

Deep mode looks at more material and traces relationships between atoms.

Response shape

Choose one of two formats:

Structured — fielded data. Good for piping into other tools or automation.
Narrative — one paragraph written for a human. Good for dropping straight into an AI agent's context.

Tips

Be specific. "auth-service rate limits" works better than "auth."
Use deep mode for high-stakes work. Writing, decisions, customer-facing responses are safer with the deeper retrieval.
Verify when conflicts surface. The briefing surfaces them; a person should still adjudicate.

Developer reference (API)

Call via API or MCP tool.

GET /api/memory/briefing?q=<topic>&depth=<quick|deep>&format=<structured|narrative>

Or via MCP:

get_briefing { query, depth?, format? }

Scope to a sphere (team/project space):

GET /api/memory/briefing?q=sports%20warehouse%20bugs&depth=deep&sphere_slug=sports-warehouse

Parameters

q / query — the question in plain language
depth — quick (default, instant) or deep (1-2s, more thorough)
format — structured (default, JSON) or narrative (prose)

Structured response example

{
  "topic": "rate limiting policy",
  "current_beliefs": [
    { "entity": "api-gateway", "relationship": "has_limit", "value": "100 req/s", "confidence": 0.91 }
  ],
  "recent_changes": [
    { "entity": "auth-service", "from": "50 req/s", "to": "200 req/s", "at": "2026-04-10T…" }
  ],
  "contradictions": [
    { "entity": "payments", "relationship": "has_limit", "values": ["20 req/s", "60 req/s"], "status": "open" }
  ],
  "authority": [
    { "person": "alice@…", "topics_owned": ["api-gateway"] }
  ],
  "gaps": [
    { "entity": "webhooks", "reason": "no atoms in last 90 days" }
  ]
}