One knowledge base. Every agent. Always up to date.
The Logbook is where your agents' knowledge lives — company context, uploaded documents, proven playbooks, and team notes. Scoped per role, inspectable per reply.
Start informed, not empty
New agents don't start from zero. The Logbook holds company context from your website, uploaded docs, and anything your team adds. Every relevant agent can draw from it immediately.
- Company context Imported from your website and enriched over time by your team.
- Uploaded documents PDFs, notes, playbooks, and internal material — available instantly.
- Team contributions Anyone can add context that makes agents more useful.
Persistent shared logbook with vector-backed semantic search, role-based read/write policies, and file attachment support.
What worked gets saved automatically
When an agent handles a task well, DeckCrew captures the approach — the plan, the tools, and the outcome. Next time, the agent starts from the proven pattern instead of starting over.
- Automatic capture Successful task patterns saved without manual effort.
- Reusable across agents A support runbook can help any support agent, not just the one that created it.
- Improves over time More tasks completed means a richer library of proven approaches.
Task runbooks capture plan/outcome/tool-run traces. Retrieval biases toward high-success patterns. Runbooks are scoped by role and task type.
Each agent reads only what it should
The Logbook is shared, but access is role-scoped. Your support agent sees customer-facing content. Your dev agent sees technical docs. No accidental cross-contamination.
- Read policies per agent Control which entries each agent can access.
- Write policies per agent Control which agents can add or modify entries.
- Public and secret entries Some knowledge is shared broadly, some is restricted.
Role-based read/write policies with public/secret visibility levels. Entries tagged by scope, source, and creation context.
See where every answer came from
Every reply your agent gives includes attribution — which Logbook entries it used, which runbooks it referenced, and whether context came from the current conversation or shared memory.
- Four memory lanes Session, local history, shared Logbook, and runbooks — each labeled per reply.
- Inspect from the Bridge View sources for any reply and verify the agent used the right information.
- Promote useful context If a reply used good session context, promote it to the Logbook with one click.
Continuity attribution with lane labels: session, local-vector, shared-memory, and runbook sources. Context inspection in Bridge with promotion workflow.
Agents suggest what's worth remembering
DeckCrew agents surface 'this looks worth keeping' suggestions without pretending to save things automatically. You decide what becomes permanent — the system stays honest.
- Conservative detection Agents flag explicit preferences and patterns, not every passing mention.
- No false claims The system corrects unsupported 'I'll remember that' statements.
- Operator-driven promotion You review candidates and promote the ones that matter.
Conservative memory-candidate detection for explicit preferences. Bounded save proposals with deterministic correction of unsupported memory claims. Operator promotion via Bridge Context panel.
Logbook FAQ