CLI

The CLI is the universal escape hatch. Any agent framework, automation system, or human can talk to PLUR via plur <command>. It’s the same engine — every other adapter is essentially a structured wrapper around what the CLI exposes.

Install

npm install -g @plur-ai/cli@0.9.4

Or run without installing:

npx @plur-ai/cli@0.9.4 <command>

Current verified release: @plur-ai/cli@0.9.4.

Bootstrap

plur init

This:

• Initialises ~/.plur/ if missing.
• Installs Claude Code hooks and the local hook binary (if Claude Code is detected).
• Writes a default config.yaml.

For project-scoped setup:

cd ~/projects/my-app
plur init --domain myapp --scope project:my-app

This writes .plur.yaml in the project root with the named domain and scope. Hooks pick it up automatically.

Command surface

plur learn   <statement>              # save an engram
plur recall  <query>                  # search engrams (BM25 + embeddings)
plur inject  <task>                   # load relevant engrams for a task
plur list                             # list all engrams
plur forget  <id-or-query>            # retire an engram
plur feedback <id> positive|negative  # rate relevance — trains recall

plur capture <summary>                # log an episode
plur timeline [--from] [--to]         # query episode history

plur status                           # health check
plur doctor                           # diagnose embedder, hooks, MCP

plur sync    [--remote <url>]         # cross-machine git sync

plur packs list
plur packs install   <pack-url-or-path>
plur packs uninstall <pack-name>
plur packs export    --scope <scope> --out <dir>

plur stores list
plur stores add      --name <n> --url <url> --token <t>

plur init                             # bootstrap, install hooks

Full command spec with flags: CLI commands reference.

Useful patterns

Pipe an LLM correction directly into PLUR

echo "Always validate user input at API boundaries" | plur learn --type behavioral --domain software.security -

- reads from stdin, so you can pipe corrections from logs, transcripts, or other agents.

Inject context into a shell-driven agent

context=$(plur inject "fix the auth bug in the login flow")
echo "$context" | your-agent --system-prompt

plur inject returns plain text (or JSON with --json) — ready to drop into a system prompt.

Bulk-import from another store

plur ingest --from ./old-memory.yaml --domain legacy

Useful for migrating between formats or seeding a new install from a teammate’s pack.

CI-friendly status check

plur status --json | jq -e '.embedder.healthy and .index.ready' >/dev/null || exit 1

Returns non-zero if anything’s broken — drop in a healthcheck script.

When the CLI is the right adapter

• Custom agent frameworks that aren’t MCP, OpenClaw, or Hermes.
• Cron jobs, GitHub Actions, post-commit hooks that should learn or recall.
• Wrapper scripts that ingest external content (Slack exports, meeting notes, ADRs).
• Anywhere a tool can shell out but isn’t worth a full plugin.

For richer integrations (lifecycle hooks, low-latency in-process calls), pick the plugin-based adapter for your runtime. The CLI is the lowest common denominator, by design.

Next: Custom (HTTP/SSE + REST) →