Hoshimi

English | 简体中文

An OpenAI- and Anthropic-compatible gateway connecting free model stations, cloud services and local models through one API and Web UI.

• One API for Chat Completions, Responses and Anthropic Messages.
• Account management with API keys, browser/device login, renewal and rotation.
• Model discovery with capabilities, reasoning menus and availability probes.
• Live usage with calls in flight, token and cache breakdowns, first-token latency and generation speed, hour by hour for any day, kept in a local SQLite database.
• A multilingual Web UI for channels, models, chat, usage and settings.

Quick start

Requires Python 3.10 or newer:

python -m pip install -U hoshimi
hoshimi serve

Open http://127.0.0.1:10600. The package includes the built Web UI; no frontend build is needed. If that port is busy, use the URL printed by the server.

OpenCode Zen and Kilo anonymous entry points are enabled by default, without a configuration file. Availability depends on upstream models, quota and service state. Add other provider accounts in Channels, and configure the gateway in Settings.

python -m hoshimi is equivalent to hoshimi. To locate the shared configuration:

hoshimi config path

For optional cloud dependencies, CLI options, configuration migration and source installation, see the installation guide and configuration guide.

First API call

List models and replace <model-id> with an ID from the response:

curl http://127.0.0.1:10600/v1/models

curl http://127.0.0.1:10600/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"<model-id>","messages":[{"role":"user","content":"hi"}]}'

If a model API key is configured, add Authorization: Bearer <model-api-key> to both requests. Chat Completions streams by default; send "stream": false for JSON.

OpenAI-compatible clients generally use http://127.0.0.1:10600/v1 as Base URL. Anthropic-compatible agents use http://127.0.0.1:10600 and a model from the same catalogue. See API integration and Anthropic agents.

Access keys

Configure two independent keys in Settings:

• Tool access key protects management APIs, settings and API documentation.
• Model API key protects external /v1/* inference requests.

Provider account credentials are configured separately in Channels. An empty gateway key disables authentication only for its own surface. For LAN access, configure the tool key and permitted client networks; see authentication.

Documentation

Documentation index

Full endpoint schemas are available at the running server's /docs and /openapi.json. They use the tool access key when configured.

License

GNU General Public License, version 3 only (GPL-3.0-only). See LICENSE.