Redactron is a prepaid gateway to frontier language models. One OpenAI-compatible endpoint, one balance, and a cost recorded against every single request — no subscription, no invoice at the end of the month.
If your code already speaks to OpenAI, it already speaks to us. Point the client at our endpoint and pass a Redactron key.
# pip install openai from openai import OpenAI client = OpenAI( base_url="https://api.redactron.ai/v1", api_key="<your Redactron key>", ) response = client.chat.completions.create( model="gpt-4.1", messages=[{"role": "user", "content": "Hello"}], stream=True, ) for chunk in response: print(chunk.choices[0].delta.content or "", end="")
Most of the engineering here is in the part nobody demos: what happens to your money when a request fails halfway.
You hold a balance. A request reserves its worst-case cost before a provider is contacted, and the reservation is released or settled when the request ends. You cannot be billed past what you funded.
Every request produces a record: the model, the tokens, the price version it was charged at, and the exact amount. Accounting is in micro-dollars, so a sub-cent request is a real number rather than a rounding error.
Issue a key per workspace or per service, grant only the scopes it needs, and revoke it from the console. A revoked key stops working at the gateway, not just in the dashboard.
If the gateway dies mid-request, the reservation is reclaimed by a worker that reconciles held money against what actually happened. Money does not sit stranded because a process restarted.
A stream that fails after the first byte is reported as a failed request, not a successful one with missing text. The terminal state and the charge always agree.
Requests carry a processing country, and routing decisions are made against a policy you can read. Pricing is in USD with no minimum spend and no annual commitment.
Sign in with Google. No sales call, no procurement form.
Top up in USD. What you fund is the ceiling on what you can spend.
The secret is shown once. Paste it into your client and send a request.