Add an AI assistant to your product in minutes.
Describe what it should talk about, get an API key, and it starts answering your users inside your app — in your voice, on your topic. Ask it something off-topic and it politely says no instead of making things up.
Prepaid credits, 1 credit = 1¢ · pay only for what the model actually uses · no subscription
Top up from $10 · no subscription
curl -X POST https://api.sulla.work/v1/chat \
-H "X-Agent-Key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "message": "Hello" }'Up and running in three steps
-
Describe your assistant.
In plain words: what it knows, how it should sound, and what it should refuse. That's the whole setup — you can change it any time.
-
Get your API key.
One key per assistant. Issue it, rotate it, or revoke it from your account.
-
Call it from your app.
One HTTP request to
POST /v1/chat— the answer streams back word by word, so your users see it as it's written.
Works with any stack. If it can make an HTTP call, it can talk to your assistant.
curl https://api.sulla.work/v1/chat \
-H "X-Agent-Key: $SULLA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"message": "Hello!"
}'How you pay, in plain language
You buy credits up front and your assistant spends them as it works. One credit is one US cent. You pay for what the AI model actually uses, at our published rate for that model — nothing more, and no subscription.
What a credit is
- 1 credit = $0.01. Top up $10, get 1,000 credits — enough for about 16,000 typical replies on gpt-5.4-mini. Actual usage varies, so treat that as a guide, not a promise.
- Each reply costs credits based on how much the model reads and writes to produce it — a short reply costs less, a long one costs more. You pay for actual use at published rates — the math is mechanical: usage times the published rate.
- Published rates at launch: gpt-5.4-mini — 0.15 credits per 1,000 input tokens · 0.90 credits per 1,000 output tokens (tokens are how the model measures text — roughly a word or part of one). The rate for every model we offer is published on this page and in your account.
- Every reply is metered, including a polite refusal of an off-topic question — the model still did the work to recognize it. Refusals are short, so they cost little.
- The built-in tools your assistant uses along the way are included in the metering. If you connect a paid third-party service of your own, that provider bills you directly.
How to know what it will cost
- You choose the model per assistant. gpt-5.4-mini is the default and the most economical; heavier models are there when you need them, at their own published rate.
- A typical reply on gpt-5.4-mini costs about 0.06 credits. Actual cost varies with the model and with the length of both question and answer — the examples here are estimates; the rates are exact.
Topping up
- Top up $10, $50, or $100 — that's 1,000, 5,000, or 10,000 credits. Prices are in USD; you can pay the same amount in EUR or CZK.
- Pay by bank transfer in CZK, EUR, or USD — credits are added as soon as the transfer arrives. Card payments become available soon.
- Credits are valid for 12 months from the day you top up.
Beyond self-serve
Below: a version where we connect your systems for you, and one that runs entirely on your infrastructure — neither is open for orders yet. Self-serve comes first.
What's in the box
Several assistants, one account.
Each with its own prompt and its own key, fully separated from the others.
Stays on topic.
You decide what the assistant may talk about; off-topic questions get a polite refusal instead of a guess.
Streaming answers.
Replies arrive word by word, so your users never stare at a spinner.
Keys you control.
Issue, rotate, and revoke API keys from your account, per assistant.
Your prompt, verbatim.
The instructions you write are stored in your account and given to the model exactly as written — no hidden rewriting.
Honest when it doesn't know.
If a question is on topic but the assistant has no source for it, it says so rather than answering something adjacent.
Start simple. Go deeper when you need to.
Same product, three levels of involvement — pick the one that fits, switch when you outgrow it.
Self-serve
Active
You set it up yourself, in minutes.
Create an account, describe your assistant, get an API key, and you're live. Run several assistants under one account, each with its own key. Pay with prepaid credits — 1 credit = 1¢, metered on what the model actually uses. Top up $10, $50, or $100.
We connect your data
Your systems, our servers.
Point us at your API, booking system, or database. We build the connection, host it, and keep it running — under NDA if you need one — so your assistant answers from your live data. Usage is metered in the same credits as always.
Not open for orders yet — self-serve comes first.
Self-hosted
Runs entirely on your infrastructure.
We deliver the whole product as code — container image, database migrations, a runbook — and help you install it. It runs on your infrastructure; the assistant calls the AI model provider of your choice — through your own account, if you prefer.
Not open for orders yet — self-serve comes first.
Where it fits
-
Support and FAQ inside your SaaS.
Answer "how do I…" questions right in the app, in your product's voice, without a support queue.
-
A client project that needs a "wow".
Agencies and freelancers add an assistant to a client's site or app and pass the cost on at a price they can predict.
-
An assistant over your own data.
Orders, bookings, inventory, internal docs — connected by us, or running on your servers.
FAQ
What is a credit, and what does one reply cost?
A credit is one US cent — top up $10, get 1,000 credits. Each reply costs credits based on how much the model reads and writes to produce it, at the published rate for the model you chose (at launch: gpt-5.4-mini, 0.15 credits per 1,000 input tokens and 0.90 per 1,000 output). A typical reply is about 0.06 credits — actual cost varies with the length of both question and answer; heavier models cost more per reply.
What happens when my balance runs out?
Your assistant stops answering and your app gets a clear signal that the balance is empty. Nothing is billed behind your back. Top up and it continues.
How long are credits valid?
Credits are valid for 12 months from the day you top up.
Which AI model powers it?
You choose, per assistant. gpt-5.4-mini is the default — fast and the most economical; heavier models are available at their own published rate. We manage the provider, the keys, and the upgrades — so you don't have to.
Can it answer off-topic questions?
No — by design. You define the topic; anything outside it gets a polite refusal. A refusal is metered like any other reply — the model still had to read the question — but it's short, so it costs very little.
Where is my data processed?
On our servers, in your own account, separated from every other customer's. If your compliance rules need more than that, ask about self-hosting.
Can I get it self-hosted?
Not yet — self-hosted delivery isn't open for orders. The self-serve product comes first; we'll update this page when that changes.
How do I pay?
Prepaid credits: top up $10, $50, or $100 by bank transfer in CZK, EUR, or USD. Prices are in USD.
Give your users an assistant that actually answers.
Create an account, paste your prompt, and send your first message in a few minutes. Prepaid credits, 1 credit = 1¢, no subscription.
Top up from $10 · no subscription
Early access — write to support@sulla.work