TraxAgent
Security

Built to be trusted with your billing data.

TraxAgent only ever reads usage and billing data, from your AI providers, coding agents and seats. Here is exactly how keys, accounts and data are protected, and how we check it.

Read-onlyUsage and billing reports only
0Prompts or outputs collected
AES-256Keys encrypted before storage
Last 4Characters of a key ever shown
Security

Your key, end to end

TraxAgent needs one thing from you: a read-only view of your bill. Here is everything that happens to the key.

Read-onlyEncrypted at restLast 4 shownNo prompts storedRevocable
  1. You paste an admin keySent over HTTPS only (TLS with HSTS). Nothing to install.
  2. Encrypted before it's storedAES-256-GCM, tied to your account. The encryption key lives apart from the database.
  3. Used only to read reportsUsage and cost reports: token counts and dollars. Shown back to you as the last 4 characters only.
  4. Gone when you say soDisconnect to delete it at once, or revoke it in your provider's console any time.

Never: reads prompts or outputs, creates keys, changes provider settings, or sits in front of your traffic. Security team can't share an admin key? Use the self-hosted collector (Integrations). Everything we do to keep it safe →

Your provider keys

Encrypted before storage

AES-256-GCM with a fresh random nonce per key, bound to your account and provider so a key can't be decrypted for anyone else. The encryption key is kept apart from the database.

Read-only use

Keys are used only to read usage and cost reports. TraxAgent never sends prompts, creates keys or changes provider settings.

Never shown again

After you connect, only the last 4 characters are shown. Keys are never written to logs or sent to your browser.

Deleted on disconnect

Disconnect removes the key at once. You can also revoke it in your provider's console any time.

Sign-in and access

No passwords to steal

Sign-in is by single-use email link, redeemed only by a button press so email scanners can't use it up. Replies are the same whether or not an account exists.

Roles enforced on the server

Owner, admin, member and viewer permissions are checked on every request, not only hidden in the page. Your data is always scoped to your own account.

Protected sessions

Session cookies are HttpOnly, Secure and SameSite. Every change needs a CSRF token tied to your session, compared in constant time.

Abuse limits

Sign-in emails, sign-ups, uploads, data ingest and the AI assistant are rate-limited. Bot sign-ups are filtered.

Application security

No SQL injection

Every database query uses bound parameters; user input never becomes SQL text.

AI prompt-injection guards

The AI expert has no tools and can't take actions. Your account data is passed as clearly marked data, never as instructions, and anything shaped like a key is masked from its answers and from emails.

Cross-site scripting

Pages escape all user and provider text. A strict Content Security Policy blocks framing, foreign scripts and form posts to other sites.

Transport and headers

HTTPS only with HSTS, nosniff, a strict referrer policy, X-Frame-Options and a locked-down permissions policy. Request sizes are capped.

Your data

No prompts or outputs

Provider reports hold token counts and dollars, not message content. Claude Code and Codex telemetry is configured without prompt text, and Cursor usage exports and usage events carry only models and token counts.

Least data

TraxAgent stores usage totals, key names and the people on your account, and nothing it doesn't need to show you spend.

Self-hosted option

Security teams that can't share an admin key can run the collector on their side and send only the usage data.

Trust center

What a security review usually asks, in one place.

What TraxAgent does with your key

  • Reads usage and cost reports
  • Lists keys, workspaces and projects by name
  • Never makes model calls or spends money with it
  • Never creates, changes or revokes keys
  • Never sees prompts or outputs

Data we keep, and never keep

  • Token counts and dollars by key, model and day
  • Key names, team labels, fix statuses
  • Prompt or response text
  • Full card numbers (Stripe holds them)

Subprocessors

ProviderWhat forNotes
RailwayApplication hosting and databaseUnited States
StripePayments and subscriptionsCard details go to Stripe, never to us
ResendSending emailsAlerts, invites, reports
AnthropicTraxAgent AI expert answers and daily model tipsSummary figures only, never prompts; not used for training
SentryError reports when something breaksNo provider keys or passwords included

The AI providers you connect (Anthropic, OpenAI) are the sources TraxAgent reads, under your own agreements with them. Full details in the Privacy Policy.

Is our data used to train AI models?

No. TraxAgent doesn't train models on your data, and Anthropic doesn't train on the API data our AI expert sends.

Does TraxAgent change anything in our AI accounts?

No. TraxAgent only calls the endpoints that read usage, cost and key lists. It never creates or changes keys, settings or workspaces, and never makes model calls with your admin key.

Do you see our prompts or model outputs?

No. Provider usage reports contain token counts and dollars, not content. The optional usage tags your code can send carry token counts and labels only.

Where is our data stored, and for how long?

In the United States, on Railway. While your account is active. Disconnecting a provider deletes its key at once; closing the account deletes the rest within 30 days, except billing records the law requires.

What happens if a key leaks from TraxAgent?

Keys are encrypted before they're stored and only decrypted to fetch reports. Where your provider offers read-only admin keys, use one. You can revoke the key in your provider's console at any time, and disconnecting it deletes it from TraxAgent.

We use an IP allowlist in the Claude Console or OpenAI. What do we add?

TraxAgent reads your usage from these fixed addresses: 162.220.232.251, 152.55.177.181, 152.55.177.193. Add them to the allowlist before you connect the key, or the nightly read is refused. If one is refused, Connections says so and names these addresses.

Start a security review

Send your questionnaire to support@traxagent.io with "Security review" in the subject. We answer within 2 business days and can walk your team through the data flow on a call.

Email us

Security reviews

Last code review: October 17, 2026. Covered: SQL injection, AI prompt injection, cross-site scripting, access control between accounts and roles, secrets handling, rate limits and security headers. Every finding was fixed or scheduled.

  • Reviews run after major changes and at least quarterly. An independent penetration test is planned before general availability.
  • Found a vulnerability? Email support@traxagent.io with "Security" in the subject. We reply within 2 business days and don't pursue good-faith research.