F1 Security Commitments

Open-source code

The Worker source that runs every F1 request is MIT-licensed and publicly available at github.com/mini-on-ai/f1. Every production deploy is tagged. Your dashboard footer shows the exact commit hash currently running.

You don't have to trust our word — you can read the code and verify what we do with your data before signing up.

What we store

Each API call creates one row in our usage_events table with:

• Timestamp
• Model ID
• Input token count, output token count, cache token counts
• Computed USD cost
• HTTP status code
• Latency (ms)
• SHA-256 hash of the first 2048 characters of your input (for detecting repeated prompts — never reversible to the original)

We never store: prompt text, response text, tool call arguments, user data in your messages, or any content from the request or response body. This is a hard policy enforced at the code level — the Worker logs are also audited to ensure no body content appears.

You can opt in to storing the first 200 characters of each input to enable the "Top expensive prompts" insight. This is off by default. Even with opt-in, full bodies are never persisted.

Your Anthropic API key

Your Anthropic key is stored in Cloudflare D1 using AES-GCM authenticated encryption:

• A master secret (F1_KEY_ENCRYPTION_MASTER) is stored as a Cloudflare Worker secret — never in code or environment files.
• For each account, a unique random 16-byte salt is generated. HKDF-SHA256 derives a 256-bit AES-GCM key specific to your account from the master + salt.
• A fresh 12-byte IV is generated for every encryption. Ciphertext, IV, and salt are stored in D1 as BLOBs.
• Decryption happens only inside the Worker runtime, only in the proxy hot path, and only to forward your request to Anthropic.
• The plaintext key is never written to logs, KV, or any other storage.

You can verify the encryption implementation in src/crypto.js.

Key-access audit log

Every time your Anthropic key is decrypted, a row is written to the key_access_log table with:

• Timestamp
• Reason (proxy_forward, key_rotation)
• A privacy-preserving hash of the source IP (SHA-256 of YYYY-MM-DD:ip, truncated to 16 hex chars — detects repeat IPs without storing raw addresses)

Your dashboard shows the last 100 entries on the "Key access log" tab. If you see entries you don't recognize, contact us immediately.

Data retention

• Usage events — kept 90 days in full, then aggregated to daily rollups and deleted.
• Opt-in prompt prefixes — same 90-day schedule.
• Key-access log — last 1000 entries retained, FIFO.
• Account + key data — retained while subscription is active, plus a 30-day grace period after cancellation.

To delete all your data immediately, use the DELETE /api/account endpoint from your dashboard. This hard-deletes all rows across all tables for your account.

No third-party analytics

There are no analytics SDKs, tracking pixels, A/B testing scripts, or third-party JavaScript on the F1 dashboard. The only external connections from the dashboard page are to the F1 Worker and to Stripe (for the billing portal). Our Content-Security-Policy enforces this.

Subprocessors

We will update this page if subprocessors change. No other parties receive any data from F1.

GDPR / DPA

F1 stores email addresses (for welcome emails and billing) and usage metadata. For Scale-tier customers who require a Data Processing Agreement, contact hello@mini-on-ai.com. A standard DPA template is available.

EU data residency (Cloudflare EU-only routing) is available for Scale tier. Request it at signup or via email.

Responsible disclosure

If you discover a security vulnerability in F1, please email security@mini-on-ai.com with a description and reproduction steps. We will acknowledge within 48 hours and aim to remediate within 90 days. We do not currently offer a bug bounty, but we will credit researchers in the changelog unless they prefer anonymity.

See also: /.well-known/security.txt

Anonymous founder, verifiable trust

mini-on-ai is operated by an anonymous brand. We acknowledge this creates a trust asymmetry for a product that handles API keys. Our mitigation strategy is deliberate:

• The Worker source is MIT-licensed and public — read it before you trust us with a key.
• Your key is encrypted at rest and the encryption logic is in the public repo.
• Every key access is logged and visible to you.
• Our subprocessors (Cloudflare, Stripe, Brevo) are recognizable infrastructure providers you can evaluate independently.
• Payment and account deletion are fully self-serve via Stripe — you are never dependent on us for those.