Privacy summary: what OpenLLM keeps

OpenLLM sits between your clients and your model providers, so it handles your prompts, your credentials, and a record of your usage. This page summarizes what happens to each in plain language. It is a summary, not the policy. The Privacy Policy and the Terms of Service are what apply.

What OpenLLM keeps

Data	What is kept	How long
Account	Your user id, email, terms acceptance, and preferences such as fallback chains	Until you delete your account
Request metadata	For each `/v1` call: model, provider, token counts, latency, status, error code, cost estimate, endpoint, `Idempotency-Key` if sent, time, and the API key id	For the life of your account. Older rows may be summarized or pruned.
Audit log	Account events such as sign-in, vault changes, credential changes, and key creation and deletion, with the source IP address	Up to 24 months
Provider credentials	Ciphertext only, encrypted in your browser	Until you delete the credential, reset your vault, or delete your account
Memories	Encrypted memory text and a plaintext search vector	Until you forget the memory, reset your vault, or delete your account
Chats	Encrypted transcripts, with plaintext titles	Until you delete the chat or your account. A vault reset makes them unreadable.
Generated media and chat attachments	Unencrypted files at public, unguessable URLs	Until you delete them. Chat attachments are also deleted with their chat or a vault reset.
Daemon diagnostic reports	Version, platform, error codes, and timings. On by default once you accept analytics cookies; turn them off in Settings → Account → Legal or on the machine.	30 days after they are sent
Idempotency cache	The response to a request that carried an `Idempotency-Key`	10 minutes

What OpenLLM doesn't keep

Prompts and responses of /v1 calls are not logged. The one exception is the 10-minute idempotency cache: when a request carries an Idempotency-Key header, its response is held for up to ten minutes so a retry returns the same answer instead of calling the provider again.
Files, images, and audio you send to /v1 endpoints exist only in memory while they are forwarded.
Your recovery phrase and vault key are never stored in readable form.
Your provider credentials are never stored in readable form. They are decrypted in memory only for the request that uses them.

No training

OpenLLM does not use your prompts, responses, files, memories, or other content to train, fine-tune, or evaluate any AI model.

Where your requests go

Your own providers. A request served by a provider you connected goes to that provider, with your credential. Its terms and privacy policy cover what it does with the request.
Included models. Requests to included openllm/ models are served on OpenLLM's own accounts with OpenRouter and Alibaba DashScope. The privacy policy makes no promise about how long these services keep data. If that matters for a request, route it to a provider you connected. See Included models.
Subscriptions and local models run on your own device through the daemon. See Devices.
Memory. Memory text is sent to AWS Bedrock to compute its search vector. The policy states that AWS does not use this content to train models.
Marketplace tools such as context.dev call their own services. Review a tool's terms before you send it data.

Where your credentials live

Provider API keys and custom endpoint keys are encrypted in your browser with your vault key before they reach OpenLLM. The server stores only ciphertext. Subscription logins stay on your own machine. For the details and the limits of this design, see Security and vault.

Delete your account

If you have an active Pro subscription or trial, cancel it first in Settings → Billing → Plan. The app asks you to, so you aren't billed after the account is gone.
Open Settings → Account → Profile and select Delete next to Delete account.
Read the warning, select Continue, and type the confirmation phrase.
Select Delete my account.

Deletion permanently removes your API keys, provider credentials, usage and audit history, memories, and preferences, and signs you out. There is no recovery. It doesn't delete the files behind your generated media and chat attachments, so delete those first if you don't want their public URLs to keep working. The policy allows OpenLLM to keep limited records where the law requires it or for fraud prevention, accounting, or disputes.

For other privacy requests, email [email protected].

Privacy