Introduction

Hisoka API

One API key. Three wire protocols. OpenAI-compatible and Anthropic-compatible, so the
client you already use works without a plugin.

Hisoka is a gateway in front of several model providers. You send one request in the
format your tool already speaks, and it is routed to a model that can answer it. You pay
from a prepaid wallet, priced per million tokens at the rate published for that model.

What you get

Every client works. Claude Code speaks Anthropic Messages, Codex CLI speaks the
OpenAI Responses API, and Cline, Roo, Continue and aider speak OpenAI chat. All three are
served. None of them is a second-class path.

Prices you can read before you spend. GET /models is open, no key needed. It
returns each model’s price per million tokens and what it accepts, so a client can show a
real cost before the first call.

A wallet, not a surprise bill. Every call settles immediately against your balance.
The response says what that call cost, in dollars and in the exact integer micro-USD the
wallet moved.

Agent-friendly by design. The catalogue publishes context_window,
max_input_tokens and capability flags, so a coding agent sizes its own budget instead
of guessing and compacting at the wrong threshold.

Pick your endpoint

Your tool Endpoint
Claude Code, Cursor, Zed POST /messages
Codex CLI POST /responses
Cline, Roo, Kilo, Continue, aider, any OpenAI SDK POST /chat/completions
A wrong endpoint is not a degraded experience

Those clients have no fallback. Point one at the wrong endpoint and it reports the
gateway as broken, not as misconfigured. Check this table before anything else.

Start here

Reading this as an agent

Fetch /llms.txt for one line per endpoint, or /llms-full.txt for every field plus a
runnable curl. Both are generated from the same specification this site is built from,
so they cannot drift from it.