plouto

Plouto plugin

AI engineering intelligence for Plouto. Workspace observability and policy push-down for AI coding agents — see what your team actually spent on Claude Code, push down a recommended model strategy, track compliance and effect.

Install

One line, cross-agent (Claude Code, Cursor, Codex):

npx plugins add PloutoAI/plouto-plugin

The runner auto-detects which agent tools you have on PATH and installs Plouto into each. Restart your agent to load.

Credentials

After install, drop your workspace credentials into the agent's environment (find them in your dashboard under Connect):

export PLOUTO_API_URL=https://api.plouto.ai
export PLOUTO_TOKEN=<your-bearer-token>

Legacy SCALENE_API_URL / SCALENE_TOKEN are honored as fallbacks.

That's it. The plugin registers SessionStart, PreToolUse, SessionEnd and StopFailure hooks. Each Claude Code session auto-syncs when it ends.

Alternative: Claude Code marketplace

If you'd rather install through Claude Code's native marketplace flow:

/plugin marketplace add PloutoAI/plouto-plugin
/plugin install plouto

Same plugin, slightly more keystrokes.

What the plugin does

Hook	What it does
SessionStart	Fetches the workspace's recommended-model strategy, writes it to .claude/settings.local.json for the next session, and flags an in-session mismatch.
PreToolUse	When the active model is off-policy, surfaces a non-blocking dialog explaining the recommendation. The user can Allow (logged as off-policy) or Deny (and /model-switch).
SessionEnd	Reads the session's JSONL and POSTs metadata (no prompt text, no tool args, no file contents) to the Plouto dashboard.
StopFailure	Records rate-limit hits so the dashboard can show Claude Code outage exposure.

There's also a /plouto slash command for setup, manual sync, audit, and dashboard links.

Privacy

The sync script runs entirely on your machine. Only metadata crosses the network:

Exported	Never exported
Session ID, timestamps	Prompt text
Token counts (input / output / cache)	Response text
Model id (e.g. claude-opus-4-7)	File contents
Tool name (e.g. Edit, Bash)	Tool inputs or outputs
Stop reason, content block types	Thinking blocks
Git branch, project path	Any conversation content

The privacy whitelist is enforced on your machine, not the server. The sync script is stdlib Python with zero dependencies. Audit it under plouto/bin/.

Telemetry

The plugin pings ${PLOUTO_API_URL}/api/ingest/event on session start, on session end (full sync), and on rate-limit stop. The pings always include the session id and timestamp; the SessionEnd path additionally posts the privacy-bounded metadata above. Disable by unsetting PLOUTO_TOKEN.

Links

• Plouto dashboard
• Platform repo
• Plugin repo