Local checks stay on your machine.
We built threat.cafe so solo builders and security teams never have to trade away source code privacy to stop malicious agent skills, MCP servers, or packages.
What leaves your machine
During a PreToolUse check in the free builder tier, threat-cafe-hook opens your local SQLite database in read-only mode. Zero network sockets are opened.
Auditable stdlib Python 3 script under Apache-2.0.
Privacy by architecture, not just policy.
Instead of sending every dependency name to a cloud API on every tool call, we ship the signed list to your disk.
Zero phone-home in the free local hook.
When your coding agent triggers threat-cafe-hook, the check queries ~/.threat.cafe/list.sqlite locally. Your prompt text, repository paths, environment variables, and internal package names never leave your workstation.
Static signed feed syncs.
When you run threat-cafe-hook sync, your client downloads three static public files over HTTPS. No query history, repo list, or machine fingerprint is attached to that request.
No cookies or third-party scripts.
The threat.cafe site runs zero advertising trackers, zero analytics pixels, and zero third-party font or icon CDNs. Lookups in the browser counter run in local JavaScript memory.
Opt-in team and MSP decision logs.
Paid Team and MSP workspaces may explicitly configure hooks to send verdict events (known_bad, no_known_bad, unknown) to /api/v1/events.php for compliance audit packs. Raw prompts and secrets are never included.
What we store when you write to us.
We collect only what is needed to send early access invites, review submitted threat findings, resolve maintainer disputes, and serve opt-in team logs.
Data categories & retention
-
Signups
Early access & Weekly Brew. When you submit your email via /api/intake.php, we store your email address, optional role, and research-call preference until you unsubscribe.
-
Reports
Findings & disputes. When you report a suspected package or file a maintainer dispute, we store the artifact identifier, evidence URL, and optional contact email for curation audit trails.
-
Abuse guard
Hourly IP hash. Form submissions compute a salted SHA-256 hash of the remote IP and current UTC hour strictly to enforce a 25-request hourly rate limit, rather than logging raw IP histories.
-
Team logs
Workspace decision ledger. For paid Team and MSP workspaces, recorded tool verdict events are retained per workspace and exportable at any time as JSON or CSV via /api/v1/audit-pack.php.
We never sell, rent, or broker personal data or workspace decision logs to third parties.
Your privacy rights
- One-reply removal. Reply to any Weekly Brew or early access email, or write to [email protected], and we remove your address within 48 hours.
- Workspace export & purge. Team and MSP administrators can export or permanently delete their workspace decision ledger at any time.
- Open inspection. Both our public dataset (/feed.json) and local hook (/hook/threat-cafe-hook) are plain text so your security team can verify every line.
Summary commitments
- No tracking cookies or cross-site pixels.
- No third-party script or font CDNs.
- No private package names sent from the free hook.
- No sale or sharing of email addresses.
Data controller & contact. Part of manav.id (TheWORKCompany). For privacy inquiries, export requests, or deletion requests, email [email protected]. Last updated: October 2026.