Any model, one API. We keep the thread.
Point your OpenAI SDK at api.helping.io. Each request carries the new message and a response id; we keep the context, you switch models whenever you want.
14 days free · then $19/month · tokens at provider price
Works withOpenAI Python SDKOpenAI Node SDKcurlAny Responses client
Models fromAnthropicOpenAIGoogle
01 — How a thread works
You send the new message.
We send the thread.
On a stateless API you store every turn and resend all of it, in each provider's own format. On
helping.io the thread lives on the server, so a request is the new message plus previous_response_id. The model reads the same context either way.
02 — Switch models
Same thread, any turn.
Change the model, not the code.
model is a field on every request. Omit it and the previous turn's
model is used. Available ids come from GET /v1/models.
03 — Pricing
$19 a month, flat.
Tokens at provider price.
One plan. No monthly request cap, no markup on tokens, no sales step. Later features arrive as plans above Launch; Launch stays as it is for the people who bought it.
04 — Governance
Rules at four scopes.
The strictest one wins.
Budgets, model policy and rate limits apply to the organization, a project, a user or one key. A refused request returns a typed code and is never charged.
05 — Before you pay
Questions developers ask first.
Why not call the providers directly?
You can, and your code stays the same if you later do. Calling direct means you store every transcript, resend it on each turn, and keep one client and one message format per provider. Through helping.io a request is the new message plus an id, the thread is held for you, and model is a field you can change on any turn.
Does my OpenAI SDK work unchanged?
Yes. Set the base URL to https://api.helping.io/v1, use your helping.io key, and call responses.create as you do today. The compatibility page lists every field the gateway refuses, accepts without applying, or handles differently.
Who can read a thread?
Only the API key that started it. Retrieve any response by id and delete the whole thread. Pass safety_identifier, which the OpenAI SDK already sends, and a thread is tied to one of your end users so their threads can be found again.
Can I change models in the middle of a thread?
Yes. model is a field on each request. Turn one can run on Anthropic and turn two on OpenAI, continuing the same thread. Omit model on a continuation and the previous turn’s model is used.
What do tokens cost?
Provider pricing, passed through with no markup. Credits are prepaid; a top-up carries a 5% + $0.30 processing fee. The admin shows your balance and a usage table per call: date, model, total tokens, credits, balance after.
What happens to the price later?
$19 a month is the founding price, locked for twelve months. Later features arrive as plans above Launch. Launch stays as it is for the people who bought it.
Is there a sales call or waitlist?
No. Signup is self-serve: create an account, create a key, make the first call. The trial is 14 days.
Start the trial and make the first call.
No sales step, demo or waitlist. Sign up, create a key, point your client at api.helping.io/v1.