Privacy — what FreeModel stores, and what it never sees

The short version

Your prompts and the model's replies are not stored. We record metadata about each request: which model, how many tokens, whether it succeeded, how long it took.

One exception: the agent endpoint described below stores the task text you send it. Everything else on this page follows from those two facts.

What is recorded

Recorded	Not recorded
The model and provider used	The text of your prompt
Prompt, completion and total token counts	The model's reply
HTTP status and request duration	Files you upload
Timestamp	Your name, unless you created an account
Token balance against your key	Anything used for advertising

Metadata exists for three reasons: routing (the router reads usage to decide what is available), billing against your quota, and diagnosing failures. It is not used for anything else.

The exception: the agent endpoint

The agent endpoint accepts a task description and a system prompt, and both are stored with the run record: the task text truncated to its first 500 characters, the system prompt up to the same length. This is the only place on the platform where request content is written to a database.

If you are sending sensitive material, use the standard chat and messages endpoints instead; those store metadata only.

How long it is kept

Request metadata is retained so that usage history stays meaningful over time. We do not currently operate an automatic expiry on it, and we would rather say that plainly than publish a retention period the system does not enforce.

You can delete your own request records at any time by calling DELETE /agent/log with your key. That removes your rows from our request log. It does not reach server-level access logs, which are kept by our hosting providers under their own policies.

Deleting your request data, step by step

Deletion is an API call rather than a form, so it can be scripted and repeated. It removes the rows tied to the key that authenticates the call.

Step 1: Take the key you made the requests with, from the console under My → API Keys. Rows belong to the key that created them, so a second key on the same account owns a different set of rows.
Step 2: Send the call: DELETE /agent/log, with your key in the Authorization header as Bearer sk-….
Step 3: Repeat for each other key on the account if you have more than one. The call is scoped to the request log and removes your rows there.
Step 4: Know the boundary. This does not reach server-level access logs, which are kept by the hosting providers listed on the sub-processors page under their own policies.

It is not an account-deletion mechanism. For a request that goes beyond what the API exposes (an account-level deletion, or a question about a record you cannot see), write to postmaster@gladec.cn rather than assuming the call covers it.

What we do not do

We do not train models on your data. There is no training or fine-tuning pipeline connected to request data. Our only outbound path for a request is the forward to the provider serving it.

We do not sell or share data with advertisers. There is no advertising or analytics SDK on this site. No third-party tracking script runs on these pages.

We do not read your prompts. They are not stored, so there is nothing to read after the fact.

Who else is involved

A request reaches a model by passing through third parties. Providers that host the models receive your prompt in order to answer it, and are bound by their own terms. Our infrastructure, authentication and email providers are listed on the sub-processors page, which names each one and says what it receives.

Your provider keys stay yours. If you bring your own key, the provider bills you directly and our role is the routing in between.

Your account

Accounts are optional — the model list and the docs need no key at all. If you create one, we hold your email address and, for third-party sign-in, the identifier that GitHub or Google returns. Registration and sign-in are protected by Cloudflare Turnstile, which sees your IP address and browser signals as part of the challenge.

Changes and contact

If this page changes in a way that affects what is stored, the change will be dated at the bottom and noted on the about page.

Questions about any of this, or a deletion request beyond what the API exposes: postmaster@gladec.cn.

Privacy FAQ

Do you store the images or files I attach?

No. Uploaded files and images are forwarded with the request and are not written to our database. For calls that carry an image the file travels the same path as the text: to the provider, and no further. Output is treated the same way, including on the 18 image models, which are 9% of the 199 listed models.

What exactly is in the request log?

One row per call: the model and provider, prompt, completion and total token counts, HTTP status, duration and a timestamp. Nothing else, and the same shape whether the call was chat (77% of the listed models are chat models) or image generation.

Does the "not stored" claim cover the agent endpoint?

No, and this page says so twice. The agent endpoint stores the task description and the system prompt, truncated to the first 500 characters each. If you are routing sensitive material, use the standard chat and messages endpoints, which record metadata only.

Does deleting my request log close my account?

No. DELETE /agent/log is scoped to the request log: it removes your rows there and is not an account-deletion mechanism. The account and its keys are a separate matter, handled at the address below.

Is my IP address in the request log?

An IP address is not one of the fields in the table above. It does appear elsewhere in the path: the bot challenge on registration and sign-in sees it, and the hosting providers keep network-level access logs under their own policies. The sub-processors page names each of them.

Are free-quota requests logged differently from paid ones?

No. Every call records the same metadata regardless of who ultimately paid for it. That is what makes routing explainable after the fact — the log is the record of which candidate answered and which were skipped, and it is not conditional on billing.