Usage

Your agent runs pk automatically. Here's what's happening and what you do.

What's automatic

On your first session, type "use pk skill" to orient your agent. After that, your harness injects a session summary automatically — you don't need to do anything.

During the session, your agent searches, creates, reads, and updates notes on its own. You don't run pk commands manually.

Your job

Two things — mostly just at the start of a project.

Seed the knowledge base

Open your AI tool and tell it about your project — decisions already made, open questions, constraints. Your agent logs them. You never touch the CLI directly.

Set up embeddings — once

Without embeddings, search only matches exact keywords. With them, search is hybrid — your agent finds notes by meaning automatically, no flags needed.

Set up embeddings →

What gets logged

Your agent picks the type. When you seed manually, pick based on what it represents.

decision A chosen direction with context and rationale.

note A stable fact or constraint.

question An unresolved uncertainty — stays open until answered.

source Raw input — a Slack thread, meeting notes, a doc.

How search works

Without embeddings configured, pk search uses keyword matching. With embeddings, it uses hybrid search — BM25 and vector results merged automatically. Your agent gets better matches with no extra effort.

No flags to add. Configure embeddings once and every search improves.