Chat completions
Standard OpenAI chat-completions shape — messages in, completion out.
OpenAI-compatible gateway
One API for every AI model you ship
Point your existing SDK at a single endpoint. Switch models with one parameter, meter every token, and keep provider keys server-side.
Prepaid token balance. No lock-in, no per-seat fee.
- Claude
- OpenAI
- GitHub
- DeepSeek
- xAI
- Mistral
- Zhipu
- Google AI Studio
- Groq
Most teams juggle a dozen provider dashboards, keys, and billing accounts just to call three models. Hisoka puts one endpoint, one key, and one balance in front of the whole catalog.
Outcome: change one line of config, call any model, and read the token spend per request in the same dashboard.
Platform
Everything between your code and the model
Stop wiring a new client for every provider. One OpenAI-compatible surface, metered and logged.
One base URL and one key format for the whole catalog. No new SDK to learn.
Switch model with a single request parameter instead of a provider migration.
Prompt and completion tokens are counted per request and drawn from one prepaid balance.
Every call keeps latency, token counts, and upstream status so failures are traceable.
Provider credentials stay server-side. Your users only ever hold their own API key.
We replaced four provider clients with one base URL and a model string.
Backend lead
Product engineering
One endpoint
One key
Metered tokens
Full request logs
One catalog, many models
Call any available model through the same endpoint. The live catalog and its pricing multipliers sit in your dashboard.
Model switching
Change the model string; keep the request body, the key, and the client.
Token accounting
Usage is metered per call against your prepaid balance, visible per request.
Playground
Try prompts against any model with temperature and token controls before shipping.
API keys
Mint and revoke keys per project. Scope what each one can reach.
Usage analytics
Requests, tokens, and success rate per model, metered in real time.
How it works
Three steps to your first call
Create a key, point your client at the endpoint, send a request.
Integrations
Drop-in for the SDK you already use.
Claude
OpenAI
GitHub
DeepSeek
xAI
Mistral
Zhipu
Google AI Studio
Groq
Frequently asked questions
-
An OpenAI-compatible gateway. One endpoint and one key in front of many models, with token usage metered against a prepaid balance.
-
Yes. Point the OpenAI SDK base URL at this gateway and keep your existing request code.
-
Prompt and completion tokens are counted per request and drawn from your balance at the model pricing multiplier shown in the dashboard.
-
Change the model string in the request body. The endpoint, key, and client stay the same.
-
No. Upstream credentials stay server-side. Your users only ever hold their own gateway key.
Swap one line. That is the migration. Create a key and send your first request.