Security
How we protect your keys, credits
and data.
Every measure here is built into the product. Where we do not have something yet, such as a SOC 2 report or two-factor sign-in, we say so.
One key, three views
primaryapi.com/dashboard/keys
- You see once
- pa_live_3Fq9••••••••••••••••When you create or rotate it
- We store
- sha256 · 9c41e0…b7d2A one-way hash, never the key
- Logs show
- pa_live_3Fq9The first 12 characters
API keys
A leaked log line never leaks a key.
Keys can spend your money, so we keep as little of them as possible and let you scope each one.
Hashed, shown once. We store a SHA-256 hash of each key, never the key. You see the secret once, when you create or rotate it.
Least-privilege access. Full, run or view. Run keys can quote and call but not manage keys or billing. View keys cannot spend.
Spend limits that hold. Daily, weekly, monthly or lifetime limits per key, checked with every charge so parallel calls cannot overshoot.
Expiry, rotation, revocation. Give a key an expiry date, replace its secret, disable or revoke it, without touching your other keys.
Prefix-only in logs. Logs identify a key by its first 12 characters. The full secret never reaches a log or an error message.
An audit trail. Creating, rotating and revoking keys, and approving an agent’s trial key, are recorded with the time, who did it and from which network address.
Accounts and sessions
Sign-in that gives little away.
The dashboard and the page where you approve an agent’s key sit behind these controls.
Passwords. Hashed with argon2id at the OWASP baseline (19 MiB of memory, two passes). Never stored or logged as typed.
Sign-in answers. Sign-in and password reset answer the same way, with the same work, whether or not an account exists.
Sessions. A random token in an HTTP-only, SameSite=Lax cookie, Secure over HTTPS. We store only its hash, and it expires after 30 days without use.
Cross-site requests. A change made with a session cookie must come from our own origin. Requests from other sites are refused.
GitHub, Google and AgentID sign-in. Optional, with a state value and PKCE protecting the redirect back to us.
Email links. Verification and reset links carry random tokens that expire. We store only their hashes.
Every call
Guard rails on every call.
Built to refuse first and ask later.
- URL inputs
- APIs that take a URL fetch it through a server-side request forgery guard. Private, loopback, link-local and cloud-metadata addresses are refused, every redirect is checked again, and response size and time are capped.
- Strict input
- Every input is checked against the API’s schema before anything runs. Unknown fields are rejected, and request bodies are limited to 2 MB.
- Rate limits
- By default 200 requests per 10 seconds per key and 1,000 per minute per account, answered with 429 and a Retry-After header.
- A cap per call
- Set
maxCreditson a call and it never costs more than you allow. - Headers
- Standard security headers on API responses. Other websites can call the public API from a browser, but never with your session.
Data handling
Keep less, for less time.
Enough to bill you correctly and let you re-read an answer, and no more. The details are in the privacy policy and the data processing addendum.
In transit. The website and the API are served over HTTPS.
Your identity. A call carries only the inputs it needs. Your API key, account email and IP address stay with us.
Payments. Card numbers go into our payment processor’s hosted checkout and never reach our servers. Credits are added once per payment, and only after the processor confirms it.
Stored results. Kept for 24 hours by default so you can re-read them for free, then purged.
Restricted APIs. For people search, contact data and other restricted APIs we keep only a hash of your input, and access needs our approval first.
What we do not have yet
No badges we have not earned.
Security pages tend to list certifications. We would rather tell you where we stand, as of October 2026.
No security certification. We do not hold SOC 2, ISO 27001 or any other certification today. We will say so here when that changes.
No external penetration test yet. Our own automated tests and code review cover these protections, not yet an independent assessment.
No two-factor sign-in yet. Until it ships, signing in with GitHub or Google lets their two-factor settings protect your account.
No paid bug bounty. We cannot offer rewards yet. We will thank, and if you want name, anyone who reports an issue responsibly.
Responsible disclosure
Found a problem? Tell us first.
Email [email protected] with “Security report” in the subject. Please do not open a public issue.
Include
- What the issue is and why it matters
- Steps to reproduce, with the endpoint or page involved
- Request IDs (
req_…) if you have them - How we can reach you
Please
- Test only with your own account and keys
- Never access, change or spend another customer’s data or money
- No denial of service, spam or social engineering
- Give us reasonable time to fix it before you publish (we suggest 90 days)
From us. We aim to acknowledge reports within three business days, keep you updated while we fix the issue, and name you publicly if you would like. We will not pursue research that follows these guidelines in good faith.
Start with a scoped key.
Give each script or agent its own run-only key with a spend limit. If one leaks, revoke it and nothing else changes.