Open models. One API. Start free.
Open-weight models, four of them cheaper than OpenRouter’s lowest listed price — Synap v1 picks for you.
Refer a friend, earn up to $500 — share your Synap link →
Works with the tools you already use — and listed on
API prompts not stored
How this compares. Most AI APIs keep your requests on their servers for a while, even when they don't train on them:
- OpenAI — retains inputs and outputs up to 30 days by default for abuse monitoring; true zero retention needs separate account approval, and not every endpoint qualifies.
- Anthropic — auto-deletes within 30 days by default; zero retention requires a signed agreement.
On Synap it isn't an add-on. Every request we send out is checked in code against a zero-retention allowlist before it leaves — a provider able to store it is refused before one ever runs.
First purchase
Up to 50% off your first month
One code, by email. It takes 50% off the first month on BYOK, 40% off the first month on Start, Pro, Developer, Max, and 25% off a one-off credit top-up.It takes 50% off the first month of a monthly plan, or 25% off a one-off credit top-up.
We use your email only to send this code.
No sign-up form between you and the first call.
- 01Try itOpen any model above and chat. No account.
- 02Get a keySign in and create your free API key. Add credits when you need paid models.
- 03Keep your codePoint any OpenAI SDK at us: change the base URL and key, pick a model.
Pay withCardTONUSDC (x402)Card from $10 · TON and USDC from $1
from openai import OpenAI
client = OpenAI(
base_url="https://synap402.com/v1",
api_key="sk-pool-...",
)
r = client.chat.completions.create(
model="qwen3-coder-30b",
messages=[{"role": "user",
"content": "Hello"}],
)
Rather not pay yet? Sign in with Google for 100,000 free API tokens a day for your first 7 days, then 50,000 a day, on qwen3-coder-30b, shared per account and network address. No card. How the free tier works
What happens to what you send.
- Not used for trainingWe do not use your prompts to train models.
We do not sell your data or use it for advertising. - API prompts not storedWe do not retain prompt or completion text for API requests.
We record token counts, the model and the time, so we can bill you. - Confidential modeOn supported models, add
"confidential": trueto a request.
It then runs inside an attested GPU Trusted Execution Environment, and the operator cannot read it even with root access. Which models
The exceptions, in plain words.
If our safety filter blocks a message in the site chat, or one is reported there, up to 120 characters of it are kept in a moderation log. Requests made with an API key are never written down, even when they are blocked.
Batch API files you upload, and the results file, are kept as your storage until you delete them. When a batch finishes, we keep no other copy of what it asked or answered.
Conversations you save in the site chat are stored in your account, and you can delete them.
Models run on machines from independent operators, who are required not to keep request content.
Read the full privacy policy →