One API key.Zero retention.
Nothing you send is retained. Ever. Not by us, and not by any model provider on our list. We list a model only when its provider is bound by a zero-retention contract. No contract, no listing.
OpenAI- and Anthropic-compatible: change the base URL, keep your code.
Early access. Public sign-up is not open yet. To get in early, email hello@zeroretentionapi.ai.
More updates soon. New models appear here when their zero-retention contract is signed.
Your prompt passes through. Nothing stays behind.
Your app sends one request to our router. The router sends it to a model provider on our list and streams the answer back. No one on that path keeps what you sent.
-
01
You send a request
Use your usual SDK. Set our base URL, your key, and a model id from your model list.
Kept: nothing -
02
We check and route it
We check your key and your balance, pick the provider for that model, and convert the format if needed. All in memory.
Kept: nothing -
03
The provider answers
The model runs at a provider on our list, under a zero-retention contract.
Kept: nothing -
04
We meter and forget
We stream the answer back, count the tokens, and bill the cost. The prompt and the answer are gone.
Kept: billing numbers only
Peace of mind, on both sides of the call.
Nothing is retained. Ever. Not by us, and not by any model provider on our list. Here is how we keep that promise.
Our side
Nothing stored- No prompts, system messages, or files. Ever.
- No model outputs, streamed or not
- No provider error text, because it can quote your input
- No training on your data, and no sale of it
The provider side
Contract- A curated list: every model is served under a zero-retention contract
- No contract, no listing. From day one.
- A legal contract, not a settings toggle or a policy page
- You never pick a provider that keeps data by mistake: those are not on the list
What we do record: billing numbers, never content
To bill you and show your usage, we record per call: which key called which model and when, the token counts and cost, timing and status, request details such as origin and user agent, and optional labels you send (for example an end-user id). Never the prompt. Never the answer.
A curated list, not a catalog of everything.
Every listed model runs under a signed zero-retention agreement. The list grows as agreements are signed.
-
The provider signs
The model provider signs a zero-retention contract with us.
-
We test the model
We test it through our router on each API format before any customer call.
-
It joins your list
Your key reads the current list, with model ids, from
GET /v1/models.
Need a model we do not list yet? Tell us which one. We will ask its provider for a zero-retention contract.
Keep the SDK you already use.
Three request formats, one key. Change the base URL, keep your code.
client = OpenAI(
base_url="https://api.zeroretentionapi.ai/v1",
api_key=os.environ["ZERO_RETENTION_API_KEY"],
)
reply = client.responses.create(
model="MODEL_ID",
input="Hello",
)
export ANTHROPIC_BASE_URL=https://api.zeroretentionapi.ai
export ANTHROPIC_AUTH_TOKEN=$ZERO_RETENTION_API_KEY
claude # Claude Code, through our endpoint
Need higher volume, or a model we don't list yet?
Tell us your monthly volume, the models you need, and what your security team must see. A person reads every email.
Straight answers.
Something missing? Email hello@zeroretentionapi.ai.
The short version
- Prompts, outputs
- never stored
- Providers
- zero-retention contract
What is Zero Retention API?
Zero Retention API is a curated LLM API: one API key for many large language models, through endpoints that speak the OpenAI and Anthropic formats. Nothing you send is retained, not by us and not by any model provider on our list. We list a model only when its provider is bound by a zero-retention contract.
What does “zero retention” mean here?
Nothing you send is retained. Ever. Not by us, and not by any model provider on our list. Your request passes through our router in memory, goes to the provider, and the answer comes back the same way. We record only billing numbers: which key, which model, token counts, cost, and timing. Never the prompt, never the answer.
How do you know the model providers keep nothing?
A contract. We list a model only when its provider is bound by a zero-retention contract with us. No contract, no listing. A settings toggle or a public policy page is not enough to get on the list.
Which models can I use?
The models on our curated list. Your key reads the current list, with model ids, from
GET /v1/models. The list grows as zero-retention agreements are signed.
Need a model we do not list yet? Email us and we will ask its provider.
Do you log my prompts?
No. A log of prompts is a copy, and a copy is retention. Our usage records hold numbers and labels, not content. Provider error messages can quote your input, so we do not store those either.
Which SDKs and tools work?
Anything that speaks the OpenAI Chat Completions, OpenAI Responses, or Anthropic Messages format. In most cases you change the base URL and the key. Claude Code works through the Anthropic format. Streaming works on all three.
How much does it cost?
Each call costs the model provider's token price, with no markup. There is no subscription. When the balance reaches zero, calls stop with HTTP 402 until you add credits.
Do you train on my data or sell it?
No. We do not keep the content, so there is nothing to train on or to sell.
Can I sign up today?
We are in early access. Public sign-up is not open yet. Email hello@zeroretentionapi.ai and we will tell you when your account can go live.
One API key. Nothing retained. Ever.
Every listed model runs under a zero-retention contract. Get a key, add credits, and change one base URL.