API keys and security
An API key starts with 1xai-, is shown once when it is created, and only its SHA-256 hash is stored. Each key can have an expiry (at least 12 hours) and a spend cap in Toman; a key that reaches its cap revokes itself. Request and response text stays next to the usage report for up to 30 days. Your IP address, User-Agent and key are never sent to a model vendor.
Last updated:
Key format and storage
- A key is
1xai-followed by 32 random bytes (as base64url). - The full key is shown once, at the moment it is created. Copy it then.
- We do not keep the key itself: only its SHA-256 hash and its first 13 characters, so that you can recognise it in the list. A lost key cannot be recovered: create a new one and revoke the old.
- Every request is checked against the database, so a key you revoke is refused from its next request on. Requests already in flight finish.
Send the key in one of three headers; see authentication. A key in the URL (?key=) is not accepted.
Creating and revoking
Keys are created and revoked in the dashboard, on the API keys page (in Persian). At creation you can set:
| Setting | Value | When left empty |
|---|---|---|
| Name | Value:Any text; shown beside each call in the usage report | When left empty:Unnamed |
| Expiry | Value:At least 12 hours | When left empty:Never expires |
| Spend cap | Value:An amount in Toman | When left empty:No cap |
Keys are created and revoked in the dashboard only. The account API reads keys and never returns a key itself.
Expiry
A key past its expiry answers 401 with code key_revoked. The expiry cannot be changed after creation: create a new key. For temporary work (a contractor, an experiment, a workshop) hand out a short-lived key, so that forgetting it costs nothing.
The spend cap: a key that revokes itself
- The full cost of every billed call is added to that key's spend, whatever paid for it (wallet, plan credit or gift credit).
- When a key's spend reaches its cap, the key is revoked at once. The next request answers
401, with codekey_cappedorkey_revoked. - The cap is checked after each call settles. So the call that crosses it, and calls in flight at the same moment, complete and are billed: the final spend can pass the cap by the cost of those calls.
- With limit alerts switched on, you get one email when a key's spend reaches 90% of its cap.
- A revoked key is not reactivated. Create a new key with a new cap.
Alerts
Alerts are set on the top-up page (in Persian). Nothing is sent until you save them yourself.
| Alert | When | Channel |
|---|---|---|
| Low balance | When:When the spendable balance falls under the threshold you set | Channel:Verified email; SMS where it is available |
| Key cap | When:When a key's spend reaches 90% of its cap | Channel:Email |
| Plan allowance | When:When 90% of a plan's weekly, monthly or API allowance is used | Channel:Email |
Each alert is sent once per crossing. The spendable balance is the spendable_toman of GET /v1/wallet/summary (see the account API).
What is logged, and for how long
Next to each usage record, the start of the request and of the response (up to 64 KB each) and, when it is longer, its last 16 KB are kept so that you can read them in your own usage report. That text is erased after 30 days. The numbers (model, tokens, cost, status and time) stay for accounting.
| Data | Kept |
|---|---|
| Model, vendor, route, tokens, cost in USD and Toman, the rate used, status, request id, time, key name | Kept:With no erasure date, for accounting and disputes |
| The start of the request and response text (up to 64 KB) and its last 16 KB | Kept:30 days |
| The audio file of a transcription, and the audio of a realtime call | Kept:Not kept |
| A generated image (base64) in a response | Kept:Removed from the logged text |
You read that text yourself in the dashboard (the call log) and with GET /v1/usage; through an API key, only the text of calls made with an API key is returned, never what was written in the chat (what each key reads). There is no switch in this release to turn body logging off for a key. The privacy policy is the full reference.
What reaches the model vendor
Your request is sent again from our server with a key of our own. Of your headers only these pass; every other header is dropped:
| Route | Headers forwarded |
|---|---|
/v1/* | Headers forwarded:Content-Type, Accept, OpenAI-Organization, OpenAI-Project, OpenAI-Beta |
/anthropic/*, /v1/messages | Headers forwarded:Content-Type, Accept, anthropic-version, anthropic-beta |
/gemini/* | Headers forwarded:Content-Type, Accept |
- Never sent: your IP address (and
X-Forwarded-Forand its relatives),User-Agent, cookies,Referer,Origin,Accept-Language, and your1xai-*key. - The request body is the one you sent, except for a short list of changes given under OpenAI-compatible endpoints (for example
store: falseon OpenAI models). - The request text goes to the vendor of the model you chose, and that vendor's own data-retention policy applies to it.
Good practice
- Create a separate, named key per application and per environment (test, production); the usage report shows each call with its key's name.
- Put a spend cap and an expiry on test keys.
- Never put a key in browser code or a mobile app. For a voice call from a browser, mint a short-lived secret (realtime voice).
- To replace a key: create a new one, update the application, then revoke the old one.
- For a team, use a company account with workspaces instead of passing one key around.
1xAi is an independent gateway. It is not affiliated with, endorsed by, or a reseller for OpenAI, Anthropic, Google, DeepSeek, xAI, Moonshot AI, Z.ai, Alibaba Qwen, MiniMax or Tencent Hunyuan. Requests are relayed upstream rather than answered by models of our own, so each provider's own usage policies apply to what their models will and will not do, and their model availability is theirs to change. Current reachability is published on the status page.