A trust signal,
not a statement.

1. Boundaries by design

• Pathrule stores memories, rules, skills, findings, working preferences and related workspace records. A local project feature may inspect files to validate a finding or assemble project metadata; that does not upload the repository source automatically.
• Content you or your agent intentionally put into a knowledge entry becomes that entry's content. Avoid storing secrets or customer data there.
• Pathrule does not train models on your content.
• Knowledge MCP tools have no general shell execution or arbitrary file-read tool. Local project-learning tools can check the source files attached to a finding; they do not expose a general filesystem browser.
• Every folder is served by one edition, Pathrule Local or Pathrule Cloud, and never by the other in its place. A Cloud folder that cannot reach Pathrule Cloud gets an error, not Local content.
• Nothing from Pathrule Cloud is ever copied into Pathrule Local, on any plan, Solo included.
• Local knowledge goes to Pathrule Cloud only when you move a folder to Cloud, after a preview of what would be copied and your separate approval. The copy never overwrites, and the Local workspace is not changed. Local activity goes only with pathrule activity import --from-local, which first says what would leave the machine.
• When a folder leaves Pathrule Cloud, or you detach it, this machine's cached copy of that Cloud workspace is removed: its hook index, session caches and rendered knowledge files. Nothing is deleted on Pathrule Cloud.
• Standalone and Local Signals are separate, opt-in runtime observations. Instrumentation is proposed and approved before code is changed. Cloud Signals is part of the plans listed on Pricing.

2. Verify it yourself

Inspect the setup and the tools your agent can use:

pathrule status
pathrule doctor
pathrule integrations list
pathrule integrations doctor
pathrule telemetry

pathrule status shows which edition serves the folder and why. pathrule telemetry exists when this machine uses Pathrule Cloud; a CLI that uses only Pathrule Local has none.

The client configuration and managed files remain readable on disk. Claude Code and Codex can use native plugins or settings hooks. Cursor, Windsurf and GitHub Copilot get Pathrule's MCP entry only, with no hooks and no repository files. pathrule eject previews project cleanup; pathrule eject --apply carries out the reviewed plan.

Any MCP client can call tools/list to inspect the connected server's tools and schemas. Local stdio and Remote MCP have different catalogs; the reference describes both.

Your agent's own model-provider traffic is separate. You can inspect Pathrule traffic with your firewall or a proxy.

3. What the local runtime touches on disk

The Pathrule home is ~/.pathrule unless you set PATHRULE_HOME. It holds local stores, session credentials when signed in, installed capabilities, hook indexes and diagnostics. Credentials and local stores are restricted to your user.

In a repository, Pathrule writes the hook configuration and knowledge files of Claude Code, and of Codex when the repository has Codex files. It preserves text outside its managed regions.

The Local Console binds to loopback, uses a one-time launch link and loads no external resources. Keep the serving terminal open while you use it; Ctrl+C stops it.

4. How the cloud is gated

Pathrule Cloud runs on Cloudflare with authenticated server-side access checks. Workspace reads and writes check the current session, membership, role and workspace state. Every Cloud operation is authorized by the server; hiding a control in the client is not an access check.

• Remote MCP tokens are bound to one approved workspace and scopes.
• Removing access, archiving or deleting a workspace revokes its grants. Restoring it does not revive old grants.
• Password reset and explicit session revocation also end Remote MCP grants.
• Refresh tokens rotate. Replaying a rotated token revokes that token family.
• Sensitive workspace lifecycle actions require fresh authentication.
• Requests, payload sizes, plan entitlements and usage are checked by the server.
• Cloud traffic uses HTTPS. Deletes retain the history needed for the documented recovery window.

5. Pathrule Web in the browser

The Cloud console manages workspaces, members, settings and billing. It calls the Cloud API with your bearer session. It does not send an administrative key to the browser and does not run coding agents.

The authenticated console is separate from the public website. Marketing analytics load only after you accept analytics in the cookie banner. Product telemetry in the managed console has its own opt-in setting.

6. The MCP tool surface

Local MCP provides path-scoped knowledge authoring and retrieval, plus project learning, refresh workflows, activity and other local helpers. Remote MCP provides context and list/read operations for memories, rules and skills, plus mutation and history operations with a write grant.

Remote MCP cannot read your filesystem, install hooks or write instruction files. It has no local personal-intelligence delivery. Its schemas are Cloud-native, so clients must inspect tools/list rather than assume local arguments also work remotely.

7. AI and telemetry

No model runs on your machine. What Pathrule learns locally, a decision you stated, a lesson a failing command taught, a preference you voiced, is read from the shape of your own words and commands, never understood by a model. Cloud knowledge indexing and semantic retrieval use Cloudflare Workers AI, and Anthropic's Claude names knowledge topics from topic keywords, folder paths and at most six item titles, never item bodies. On a Cloud workspace, the Cloud model tier reads the prompts Pathrule's phrase rules found nothing in and keeps a decision you meant to last, in your words; it is paid from AI Credits. Model-assisted Cloud features process the knowledge or evidence needed for that feature; availability follows Pricing. Your coding agent's own model calls stay with its provider.

Pathrule's CLI, MCP server and VS Code extension send no usage data. No build Pathrule distributes carries an analytics key, so there is nothing to turn on, and nothing about your commands, code, knowledge or account leaves your machine for analytics. Pathrule Cloud counts its own use on its servers, only for accounts whose holder opted in on the account page. On a machine that uses Pathrule Cloud, pathrule telemetry confirms it:

pathrule telemetry

Telemetry is always off in CI. Managed product telemetry uses PostHog's EU ingestion service, separately from Cloud knowledge storage and website analytics. See Privacy Policy and Cookie Policy.

8. Distribution and updates

Install or update the CLI with npm i -g @pathrule/cli@latest, or update it with pathrule update. The local foundation is Apache-2.0; Pathrule Cloud is proprietary. See Open source.

The CLI checks for a newer release once a day. Under the default automatic update policy, an npm global install also installs a newer release by itself when no Pathrule MCP server or host is running, after checking the package against the signed release manifest. pathrule config set update.policy notify makes it only tell you, PATHRULE_DISABLE_UPDATES=1 turns updates off, and NO_UPDATE_CHECK=1 turns off the background check. The VS Code extension requires a trusted workspace before writing integration files and starting its local runtime.

Enterprise self-hosted deployments, with SSO and audit logs, are arranged through Sales; see the self-hosted reference. They are separate from the local CLI and Local Console.

9. Reporting a security issue

Email security@pathrule.io with the affected component, reproduction steps, the impact you observed, and the platform and version. pathrule --version reports the CLI version.

We aim to acknowledge reports within three business days and keep you updated as we investigate. Please test only with accounts and data you own.

10. Contact

• Security: security@pathrule.io
• General: hello@pathrule.io
• Privacy and data requests: privacy@pathrule.io