Engagements

An engagement is one saved conversation — its full history, the findings ledger, and the captured tool output behind it. Engagements persist automatically, so you can close Kerkov and pick up exactly where you left off.

How they're stored

Each engagement is a JSONL transcript under ~/.kerkov/engagements/, written as you work. Alongside it, each engagement tree (a conversation plus any agents it spawned) shares one SQLite knowledge base at ~/.kerkov/engagements/<root-id>.kb.sqlite. Everything is local; nothing is uploaded.

Resuming

From the shell:

# Resume the most recent engagement
uv run kerkov --continue

# Resume a specific one by id
uv run kerkov --resume <engagement-id>

From inside a session, /resume with no argument opens a picker of your saved conversations; pass an id to restore it directly:

kerkov › /resume
kerkov › /resume abc123

When you exit with /q, Kerkov prints the exact resume command for the engagement you were in.

Listing

/engagements lists your saved conversations. Add agents to see the agent sub-runs behind them instead, with their parent linkage and task:

kerkov › /engagements
kerkov › /engagements agents

Agent sub-runs are read-only

A delegated agent gets its own engagement id, recorded under its parent. These appear in /engagements agents, and you can open one to inspect it — but it opens read-only: resuming an agent run and continuing it would detach it from the report its parent is waiting on. Use it to review what an agent did, not to keep it going. For live agents, see the agent subsystem.

What resuming restores

Restoring an engagement brings back its conversation history and, with the knowledge base enabled, its findings ledger and recallable tool output. The mode, provider, and model in effect are the session's current ones — resuming does not force the old posture back on you.