project · status · eval · view

pdsa project

Manage the active project (each project = a separate graph DB). See Multi-project.

pdsa project set <name>   # persist the active project
pdsa project list         # projects + cycle counts (* = active)
pdsa project show         # active project / DB path
pdsa project clear        # unset (fall back to cwd name)

pdsa status

Recent cycles and the expectation hit-rate (recall).

pdsa status [--project <name>] [--limit 5] [--full] [--json]

Shows accumulated cycle count, the hit-rate (met / cycles-with-a-verdict), and a summary of recent cycles with their per-step text and verdicts. Per-step prose is truncated to one line by default — add --full for the untruncated text, or --json for the complete structured record.

pdsa eval

Per-cycle breakdown: expected / verdict / actual, plus the hit-rate.

pdsa eval [--project <name>] [--limit 10] [--full] [--json]

Use it to audit why the hit-rate is what it is — which expectations were met, partially met, or missed. --full disables truncation; --json returns the per-cycle records as data.

pdsa recall

Read accumulated learnings back out of graph memory — the same context plan injects automatically. Give a topic keyword to filter to related cycles, or omit it for the most recent.

pdsa recall ["<topic>"] [--limit 5] [--json] [--project <name>]

Returns each matching cycle’s expected / verdict / actual and the Study/Act learning narratives (full, untruncated). An agent can call this before planning to pull relevant prior context on demand. See For AI Agents.

Note

Don’t confuse this with recall the metric (expectation hit-rate). The recall command reads learnings back; the hit-rate is met / cycles-with-a-verdict.

pdsa view

Open the local, self-contained graph viewer. See Graph Viewer.

pdsa view

Serves the recorded graph in-process on a local port (works from the AOT single-file install). Study nodes are colored by verdict; REINFORCES edges and a hit-rate badge are shown.