• 1Getting started
• 2The graph
• 3Composites & _meta
• 4Dashboard & ops

Dashboard & operations

The last stop: the human-facing side. The dashboard, diagnostics, the watcher that keeps the index fresh, multi-project setups, and HTTP mode for when an agent isn't an editor.

The dashboard

pincher web resolves (or auto-starts) the local HTTP server and opens the dashboard in your browser. Tabs cover the picture at a glance:

• Overview — token-savings ROI for the session and all-time, plus the PreToolUse hook panel.
• Insights — per-tool call breakdown, complexity-tier mix, response-payload sizes, benchmark history.
• Doctor — index health at a glance: schema, DB size, advisories, a per-project table, extraction failures by reason.
• Search / ADR — drive a search or browse the project's stored decisions, right in the browser.

pincher doctor

pincher doctor is the command-line diagnostic — the same data the dashboard's Doctor tab shows. It reports the schema version, the running binary, DB and WAL size, per-project rows (so you can spot a project indexed by an old binary), advisories for pathological states (a giant DB, a "ghost" project that indexed but extracted almost nothing), and recent extraction failures.

pincher doctor
pincher doctor --json    # same data, machine-readable

The background watcher

Once an editor is wired up, pincher runs a file watcher: it polls indexed projects (fast while active, slow while idle) and re-indexes changed files incrementally — only what changed, not the whole tree. You do not run pincher index by hand in normal use; the watcher keeps the graph current as you edit.

Multiple projects

One pincher install indexes many repositories — the database holds them all, each scoped by project. Manage them from the CLI:

pincher project list           # every indexed project
pincher project rm <name>       # drop one
pincher project prune-stale     # drop old-schema, long-untouched projects
pincher vacuum                  # reclaim disk after removals

HTTP mode

Not every agent is an editor. Started with --http :PORT, pincher also serves a REST gateway — every tool is reachable as POST /v1/<tool>, with OpenAPI contracts, optional bearer auth (--http-key), and request IDs. This is the path a router or a CI job uses to call pincher without spawning a stdio child.

Keeping the index healthy

• Staleness — doctor and the health tool report how long since each project was indexed; a stale project is the first thing to check if answers look wrong.
• Drift — after a binary upgrade pincher re-indexes affected projects automatically so resolution rules stay consistent.
• verify — pincher verify re-hashes every indexed file and reports drift vs. the stored hash.