# Data and privacy > What Clonvo stores, opt-in retention and free-model data. This page explains what Clonvo keeps about your requests and the settings that control it. Settings live under **Settings → Privacy** in the dashboard and apply to the whole organization; only owners and admins can change them. ## What Clonvo stores by default For every request, Clonvo records metadata so it can bill you and show your activity: - when it happened, the organization, project and API key, and whether it came from the API or the dashboard playground; - the model you asked for, the model and provider that served it, and whether it ran on credits, your own key or a free model; - token counts (input, output, reasoning, cached and cache-write), the number of tool calls, the cost at list price and the amount charged; - latency, time to first token, status, finish reason and the number of attempts; - the error code and message when a request fails; - your app’s `HTTP-Referer` and `X-Title` headers, if you send them. **Full prompts and responses are not saved as request content by default.** They pass through Clonvo to the provider and back. Operational metadata can include provider error messages, which may contain fragments of submitted content. Gemini tool-call thought signatures are cached separately for up to 24 hours to support multi-turn requests, even when full-content retention is off. Request metadata currently has no automatic expiry. For deletion requests and the records that may need to be retained, see [Retention and deletion](https://clonvo.com/privacy.md#retention). ## Opt-in retention Turn on **Store prompts and responses** to keep request and response bodies for your organization’s requests, including failed ones. For [embeddings](https://clonvo.com/docs/embeddings.md), that’s the request with its input texts; the returned vectors are never stored. Stored bodies are kept in object storage, separate from the metadata database. Anyone in your organization who can see the request in **Activity** can open its prompt and response there. Stored bodies currently have no automatic expiry, and the dashboard doesn’t have a way to delete them. Request content deletion by emailing [support@clonvo.com](mailto:support@clonvo.com) from your account email, identifying the organization and requests concerned. We may need to verify that you have authority over the data. Turning retention off stops storing new bodies. It doesn’t delete bodies stored while it was on. These bodies are for your organization’s own use. Clonvo doesn’t use them to train or evaluate models; that only applies to [free-model data](#free-models-and-your-data) and [web tool queries](#web-tool-queries). ## Free models and your data Free-model requests are the only model requests Clonvo stores for its own use. **Use free models** is off for new organizations, and while it’s off Clonvo doesn’t send your requests to free-model endpoints. To turn it on, an owner or admin accepts these terms for the whole organization: - Free models are run by third-party providers through OpenRouter. Those providers may log your prompts and responses and use them to train their models. - Clonvo stores the prompts and responses of your free-model requests for up to 12 months. - Clonvo may use them to train and evaluate AI models, and may license them to AI developers for the same purposes, with personal data removed first. - Some paid models have a cheaper data-sharing version, today Meta’s Muse Spark 1.3 and 1.2. Clonvo uses it for your requests to those models at its lower price, and Meta may use those prompts and responses to train its models. Clonvo doesn’t store them. To opt out for a single request, send provider.data_collection: "deny". - Clonvo stores only free-model requests for its own use. Requests to paid models, including their data-sharing versions, and requests on your own provider keys aren’t stored for these purposes. - Turning free models off stops new storage and the use of data-sharing versions. To delete what’s already stored, email [support@clonvo.com](mailto:support@clonvo.com). Once they’re accepted, Clonvo stores each successful free-model request: the request as sent (without its `user` and `metadata` fields or the `name` on messages), the response, the models used, token counts and the time. Each record is deleted automatically up to 12 months after the request. The [privacy policy](https://clonvo.com/privacy.md#free-model-data) has the details. - **Paid and own-key model requests**, embeddings included, are never stored for these purposes, whatever your settings. Web tool calls are covered [below](#web-tool-queries). - **Data-sharing prices.** With the terms accepted, paid models that have a data-sharing version (today Meta’s Muse Spark 1.3 and 1.2) run on it at its much lower price, and that provider may train on those requests. Clonvo doesn’t store them. See [Meta](https://clonvo.com/docs/meta.md#data-sharing-prices). - **Turning free models off** stops new storage. To delete what’s already stored, email [support@clonvo.com](mailto:support@clonvo.com). Deleting an organization deletes its free-model data too. - **When the terms change**, free models are paused for your organization. Requests get `403` with the code `free_model_terms_required` until an owner or admin accepts the new version in **Settings → Privacy**. With free models on, you can still exclude free models and data-sharing versions for a single request with: ```json { "provider": { "data_collection": "deny" } } ``` ## Web tool queries When your agents call Clonvo’s [web tools](https://clonvo.com/docs/tools.md), Clonvo stores the inputs of each successful call for its own use: the `web_search` query and its options (such as domains, time range and `max_results`), or the URL a `web_scrape`, `web_map`, `web_extract`, `web_crawl` or `find_similar` call used and its options, plus the tool, the provider that served it and the time. Free text you write for a tool is never stored for this: not a `web_answer` question, a `web_extract` prompt or schema, `web_crawl` instructions, a `web_research` brief or a `browser_task` goal. Nor are company and people lookups: the `company_*` and `people_*` tools, and `web_search` with category `people` or `company`. It doesn’t store the results or page content, any prompts or responses, or which API key, project or user made the call. Clonvo uses these queries to train and improve its search, and doesn’t license them to others. - **On by default** for every organization, whether calls run on your own provider key or on Clonvo credits. - **To turn it off**, an owner or admin switches off **Improve Clonvo with your web searches** under **Settings → Privacy**. New calls stop being stored within about 15 seconds. Turning it off doesn’t delete what’s already stored. - **Retention.** Each record is deleted automatically up to 12 months after the call. To delete stored queries sooner, email [support@clonvo.com](mailto:support@clonvo.com). Deleting an organization deletes them too. - **Keep secrets out of queries and URLs.** Don’t put personal or confidential information, credentials or access tokens in search queries or the URLs you scrape. The [privacy policy](https://clonvo.com/privacy.md#web-tool-queries) and [terms](https://clonvo.com/terms.md#web-tool-queries) have the details. ## What providers receive Each request is sent to the model’s provider, which handles it under its own API terms and data policy. Free-model requests may pass through OpenRouter as well as the serving provider. Each [model page](https://clonvo.com/models.md) links to the provider’s policy. With [your own keys](https://clonvo.com/docs/byok.md), your agreement with the provider applies. The free-models setting is a routing control; it does not guarantee that providers keep no operational or abuse-monitoring records. ## Your account data How Clonvo handles account and billing data is described in the [privacy policy](https://clonvo.com/privacy.md).