MorseSign inStart freeOpen Morse

Is OpenRouter private? What it logs, what its providers keep, and what ZDR still allows

The Morse teamPublished 11 min read

OpenRouter is private from OpenRouter itself by default, but not from the model providers it sends your prompts to. As of November 2026, OpenRouter doesn't store your prompts or responses unless you opt in, keeps metadata on every request, and hands each prompt to a provider whose own policy decides whether it's kept or trained on.

We read OpenRouter's privacy policy and terms (both last updated August 31, 2026), its privacy and zero data retention docs, its FAQ, and the model pages for every free model it listed on November 14, 2026. OpenRouter deserves credit for how much it publishes. It labels the data policy of each endpoint, lets you enforce zero data retention on a single request, and assumes the worst about any provider whose policy it can't confirm.

OpenRouter's own pages answer the common questions this way, as read on November 14, 2026:

QuestionOpenRouter's answerSource
Does OpenRouter store prompts?No, unless you opt in, the endpoint uses Safety Retention, or content is flagged as potential CSAMData collection
What does it log on every request?Metadata: timestamps, model, token counts, and latencyFAQ, Data collection
Does OpenRouter train on prompts?No, it "does not use your Inputs or Outputs for model training"Privacy policy
Do its providers train?Some do, under their own policiesProvider logging
What does logging for a discount cost?1% off model usage, for a perpetual license to your promptsTerms
How long are private logs kept?At least 3 monthsInput and output logging
Does ZDR mean nothing is kept?Not by default on self-serve accounts, which allow Safety RetentionZDR guide
Are free models private?1 of 14 free models has an endpoint on the ZDR listZDR endpoint list

Does OpenRouter log your prompts?

No, not by default. OpenRouter says it "does not store your prompts or responses, unless you opt in" to one of two settings, or the content is flagged as potential CSAM (OpenRouter). Safety Retention endpoints are a third exception, where "OpenRouter or the provider may retain prompts and responses from any request," only to prevent abuse. Its FAQ adds that it does "zero logging of your prompts/completions, even if an error occurs" (OpenRouter).

It does log metadata on every request. The FAQ lists "timestamps, model used, token counts" (OpenRouter), and the data collection page adds latency (OpenRouter). That metadata feeds your own logs page and OpenRouter's public model rankings.

OpenRouter also reads a sample of prompts. It says it "samples a small number of prompts for categorization to power our reporting and model ranking," and that unless you've opted in to sharing, the categorization is "stored completely anonymously" (OpenRouter).

In an announcement dated October 27, 2026, OpenRouter said it will screen all inference requests for CSAM "as they pass through our service," and that flagged content "may be retained for human review." The screening covers zero data retention traffic too.

What do you give up for OpenRouter's 1% logging discount?

A broad license to your prompts. The setting OpenRouter calls "OpenRouter Use of Inputs/Outputs" gives you "a 1% discount on all model usage" in exchange for letting it use your prompts and completions to improve the product (OpenRouter). It's off by default, under Privacy settings.

The terms spell out what you grant once it's on. Section 6.2 gives OpenRouter a "worldwide, perpetual, irrevocable" license to your content "for our own commercial and business purposes," including permission to "license or sell your User Content in anonymized form" (OpenRouter). Opting in also turns on chat logging automatically. OpenRouter's data collection page says it "has never sold or licensed prompt data" (OpenRouter), so by its own account the right hasn't been used.

Private logging is a separate switch, under Observability. Input & Output Logging saves your prompts so you can review them, and OpenRouter says it doesn't use them for training, analytics, "or any other purpose" (OpenRouter). The logs sit in a Google Cloud Storage project, encrypted at rest, and are kept "for a minimum of 3 months" and longer "at OpenRouter's discretion unless you request deletion." On an organization account, org admins can read every member's logged prompts.

The privacy policy adds one more exception: "for purposes of de-bugging, we may re-identify your user ID with any stored Inputs" (OpenRouter).

Do OpenRouter's providers train on your prompts?

Some do, and OpenRouter can't stop them once the prompt arrives. Its privacy policy says "OpenRouter cannot control Model Provider-side training once user data is transmitted to a training-permitted Model Provider" (OpenRouter).

It does control where your prompt goes. Your account has a setting for whether to allow "routing to providers that may train on your data," with "separate settings for paid and free models," and if you opt out, "OpenRouter will not route to providers that train" (OpenRouter). Where it can't pin down a provider's policy, OpenRouter "assume[s] that the endpoint both retains and trains on data" (OpenRouter).

The API sets its own default per request. The data_collection field defaults to allow, which OpenRouter defines as allowing "providers which store user data non-transiently and may train on it" (OpenRouter). Setting it to deny limits a request to "providers which do not collect user data." OpenRouter calls its policy tags "our best knowledge," not "a definitive source."

If you reach OpenRouter through Janitor AI, the training setting is the one Janitor's error guide tells you to switch on. We cover that trade, and how to avoid it, in Janitor AI alternatives.

Does OpenRouter zero data retention mean nothing is kept?

Not on a self-serve account's defaults. OpenRouter now sorts every endpoint into three tiers: ZDR, "no routine storage after processing"; Safety Retention, where OpenRouter or a provider "may retain prompts and responses from any request, whether flagged or not, only to detect, investigate, and prevent abuse"; and Retention and Training (OpenRouter).

When you turn on ZDR, you choose ZDR only or Allow Safety Retention. If you haven't chosen, "self-serve accounts allow Safety Retention and enterprise accounts use ZDR only" (OpenRouter). A zdr: true flag on a request follows the same choice. On November 14, 2026, 68 of the 1,108 entries in OpenRouter's ZDR endpoint list were marked as Safety Retention.

Google models are the biggest case. Since October 18, 2026, Google requires Safety Retention for its models served through Vertex, which means it "may retain prompts that its safety classifiers flag, for up to 90 days" (OpenRouter). OpenRouter says those endpoints "don't meet a strict Zero Data Retention requirement," and choosing ZDR only makes "most Google models" unavailable.

Three more limits sit in the same guide:

  • Tools aren't covered. ZDR "only applies to provider routing for inference requests," not to plugins and tools such as web search (OpenRouter).
  • Caching doesn't count. OpenRouter has "taken the stance that in-memory caching of prompts is not considered 'retaining' data."
  • Private endpoints are taken on trust. For a deployment you connect yourself, OpenRouter records your ZDR declaration "and does not independently verify" it.

What the term means across providers is in what is zero data retention, and the abuse logs behind Safety Retention are explained in what is abuse monitoring.

Are OpenRouter's free models private?

Mostly not. On November 14, 2026, OpenRouter's models list had 14 models with a :free version, and each free version ran on a single host. We read the data policy that each model's own page shows for its free endpoint. Seven run on a host OpenRouter marks as training on prompts, 13 on a host that keeps them, and one appears on the ZDR list.

Free model on OpenRouterFree hostMay train on promptsKeeps promptsOn the ZDR list
Nemotron 3.5 Lightning, Nemotron 3.5 Super VL, Nemotron 3 Ultra, Nemotron 3 Nano OmniNVIDIAYesYesNo
Inkling 2 SmallThinking MachinesYesYesNo
LFM3-1.7BLiquidYesYesNo
Particle-1Particle AIYesYesNo
Gemma 4 26B A4B, Gemma 4 31BGoogle AI StudioNoYes, 55 daysNo
North Mini CodeCohereNoYes, 30 daysNo
ERNIE 5.5 Flash PreviewBaiduNoYesNo
FlintRecursiveNoYesNo
Dots3-Note PreviewAtlasCloudNoYesNo
Ling 3.2 FlashAnt DigitalNoNoYes

Paid versions often run on stricter hosts. Nemotron 3.5 Lightning's page, for example, lists three paid hosts marked as keeping no prompts, so the free version is the less private way to use that model. OpenRouter's pricing page lists data policy-based routing as not included on its Free plan, which allows 50 requests a day.

Stealth models carry their own terms. OpenRouter's Stealth Program terms say that if a stealth model's listing says it trains, "your User Content will be logged in full and retained by the Stealth Provider," and OpenRouter "may not, in certain instances, disclose the name or origin" of that provider.

What does OpenRouter know about you?

Your account, your device, and how you use the site. OpenRouter's privacy policy lists your IP address, operating system, browser type, and time zone setting among what it collects automatically. It uses Google Analytics, and PostHog, "a web and products analytics platform that may also utilize session replay features." It also lists "Delivering interest-based ads" among its purposes, which you can switch off in your account's advertising preferences.

Payments add more. OpenRouter takes major credit cards, AliPay, and USDC, and charges a 5.5% fee ($0.80 minimum) when you buy credits, or 5% when you pay in crypto (OpenRouter).

The policy lets OpenRouter share data with "a buyer or other successor," including in bankruptcy. It also says OpenRouter "will preserve and disclose any and all information to law enforcement agencies or others if required to do so by law," or where it believes in good faith that it's needed (OpenRouter). Your account, request metadata, and billing records are on file whatever your logging settings.

Your API keys are your responsibility. Section 3.2 of the terms makes you responsible "for all activity and charges under its account or API Credentials, whether or not authorized by you," and asks you to "promptly notify" OpenRouter of a compromised key (OpenRouter).

Media and files are handled separately. The privacy policy says images, audio, and video sent with a request aren't kept "beyond the duration necessary to route the request, except as required for abuse detection, security, billing, or legal compliance," while files uploaded through its Files API stay "until you delete them or close your account" (OpenRouter). The same policy says your data "may be transferred to our servers in the US."

Requests can stay in Europe on the Business and Enterprise plans (OpenRouter). Requests sent to eu.openrouter.ai go only to provider endpoints in Europe, though "account and billing data may be stored or processed in the US" (OpenRouter).

The chatroom keeps your history on your device. OpenRouter's FAQ says the same data rules apply to it as to the API, that its conversations are "stored locally on your device" without syncing across devices, and that you can export and import them from the chatroom's settings menu (OpenRouter).

How do you delete your OpenRouter data?

From your account settings. OpenRouter's FAQ says to open Settings, click Manage Account, and use the delete option under the Security tab, and warns that unused credits are lost (OpenRouter). The privacy policy also takes requests by email, asks you to confirm, and says the process "cannot be canceled, undone, withdrawn, or reversed" (OpenRouter).

Deletion covers "all associated personal data," unless OpenRouter is "required to maintain it for regulatory and compliance purposes or for a legal or business necessity." Logged prompts from Input & Output Logging are deleted on request to OpenRouter's support address (OpenRouter).

How do you make OpenRouter more private?

Set it up once, then check each new key and model:

  1. Turn off training for both paid and free models in your privacy settings.
  2. Enforce ZDR and pick ZDR only, for each model group you use, so Safety Retention endpoints are excluded (OpenRouter).
  3. Leave the 1% discount and Input & Output Logging off.
  4. Send provider: { "zdr": true, "data_collection": "deny" } on API requests that matter, since the API default is allow.
  5. Skip free and stealth models for anything personal.
  6. Use the Sensitive Info Guardrail to redact or block email addresses, phone numbers, and card numbers before a request reaches the provider (OpenRouter).

None of this hides who you are from OpenRouter. If you also call model makers' APIs directly, their own terms are compared in is the OpenAI API private.

How Morse keeps model hosts from keeping your prompts

Every OpenRouter request runs under your own OpenRouter account, tied to your email, payment, and IP address, and lands with a provider whose own policy decides what's kept. Staying private means keeping a set of switches right on every key and every model. Morse is a web app built so you know the arrangement with the host before you type, with a privacy mode you pick for each chat.

  • Private, for models whose hosts have agreed to keep nothing. It's the default for a new chat. DeepSeek, Kimi, GLM, and Grok run under the zero data retention agreements Morse has with its inference, search, and other providers, so they've agreed to keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't.
  • E2EE, for prompts you won't leave to anyone's declaration. Your browser checks a secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and Morse's servers relay ciphertext they can't read. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. E2EE runs open-weight models with no web search or image generation, and comes with Pro at $19 a month.
  • Anonymous, for GPT, Claude, and Gemini without an account in your name. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account. The provider sees the message without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy, so leave out details that point to you.

Morse's servers don't keep your messages or log what you write. Morse doesn't train AI on your content, sell your data, or show ads, and it runs no advertising trackers or third-party analytics. Your chat history is stored in your browser, encrypted, and because Morse doesn't store chats, it usually has very little to give authorities. Clearing your browser deletes those chats for good, and Morse can't recover them, so keep a copy of anything you need.

You can try a chat from the home page without an account, or start free with no card and 15 messages a day.

Common questions

Is OpenRouter's chat app private?

OpenRouter's FAQ says the chatroom follows the same data rules as the API and that its conversations are stored locally on your device, with no sync across devices. You can export and import them from the chatroom's settings menu.

Does OpenRouter keep images, audio, or files I send?

Not images, audio, or video sent with a request: its privacy policy says it doesn't keep them beyond routing, except for abuse detection, security, billing, or legal compliance. Files you upload to its Files API are a different case, kept until you delete them or close your account.

Is it safe to use an OpenRouter API key in other apps?

Treat it like a password. OpenRouter's terms make you responsible for all activity and charges under your keys, whether or not you authorized them, and ask you to report a leaked key promptly.

Where does OpenRouter process my data?

OpenRouter, Inc. says your data may be transferred to its servers in the US. On its Business and Enterprise plans, requests sent to eu.openrouter.ai stay with providers in Europe, though account and billing data may still be stored or processed in the US.

Ask what you actually want to ask.

80+ models, characters, and images, with every chat encrypted and kept on your device. Free to start.

Start a private chat

Read next