Skip to content

cli.md

Latest commit

 

History

History
51 lines (33 loc) · 3.29 KB

cli.md

File metadata and controls

51 lines (33 loc) · 3.29 KB
Raw

Plex CLI

You do not need the CLI for normal use. The plugin runs the reviewer (agent + MCP), /plex:analyze seeds knowledge from your PR history, and plex init sets your embedding key. The CLI is the same engine for the few jobs that aren't the review itself: setup, building the code graph, the visualization UI, and background maintenance.

There is no plex review or plex analyze command. The review and the history analysis both run through your agent, in its own fresh context (that is the whole point), so the CLI never does the reasoning — use /plex:review and /plex:analyze.

Running it

Two ways, same engine. Run them from inside your repo; every command defaults to the current git repo.

  • No install. Prefix any command with npx @sshanzel/plex. npx fetches the package once and caches it:

    npx @sshanzel/plex index

  • Global install (for a bare plex). If you run these often, put it on your PATH:

    npm install -g @sshanzel/plex
plex index

The examples below use the bare plex. Without a global install, prefix with npx @sshanzel/plex (e.g. npx @sshanzel/plex serve).

Commands

Command What it does
plex init Interactive setup: set the embedding key, then offer to index the current repo. The most common reason to reach for the CLI.
plex index [--incremental] Build or refresh the code graph. --incremental re-reads only the files that changed. (Reviews auto-index on first use and refresh on drift, so you rarely need this by hand.)
plex serve [--port N] [--stop] [--status] Start the local web UI to explore the code graph, PR brain & knowledge (http://127.0.0.1:2288). On-demand; opens Kùzu per request, so it never blocks a review.
plex sweep Run background maintenance once: close landed review loops, refresh main's graph, and consolidate knowledge decay (ADR-43). Normally auto-spawned by a review; this is the manual trigger.

Everything else happens through your agent: it reviews with /plex:review, seeds knowledge with /plex:analyze, and calls the MCP tools (index_repo, analyze_scan / add_pitfalls, reconcile_outcomes, record_outcome, consolidate_knowledge, doctor, …) directly during a review — so you don't touch the CLI for them.

Storage

Per-repo data lives outside your repo, at ~/.plex/repos/<id>/, so there is nothing to add to .gitignore. The global knowledge base is at ~/.plex/knowledge/, and your key and settings are in ~/.plex/config.json. See the Embeddings section of the README for providers and the switch-provider caveat.

From source (contributors)

pnpm install && pnpm build         # builds dist/plex.js and dist/plex-mcp.js (run under node; ADR-19)
node dist/plex.js index            # build the graph for the current repo
node dist/plex.js serve            # open the visualization UI

Heads-up: don't test the published package with npx @sshanzel/plex … from inside this repo. The repo's own package.json is named @sshanzel/plex, so npx resolves to the local copy (whose bin isn't linked) and you get sh: plex: command not found. Use node dist/plex.js … here, or run the npx command from any other directory.

See AGENTS.md for the full contributor workflow.