Krasis documentation

Connecting your apps

Use Krasis from a chat app, coding agent or local management dashboard.

Connect an OpenAI-compatible app

Start the model server and wait until it is ready. In your app, choose an OpenAI-compatible or custom provider and enter:

127.0.0.1 means the computer running the app. If the app runs on another machine, use the Krasis machine’s reachable address and server port. A container also has its own loopback address.

For local-only serving, bind Krasis to 127.0.0.1. The server’s default bind address is 0.0.0.0, which listens on available network interfaces. Set up API keys when allowing other machines to connect; use a trusted network or encrypted tunnel to protect traffic.

Find the model ID

From a terminal on the server machine:

curl http://127.0.0.1:8012/v1/models

If authentication is enabled, add -H "Authorization: Bearer $KRASIS_API_KEY" after setting that environment variable to your actual key. Use the returned model ID in your app or request. The API reference shows a complete request.

Use a coding agent

Point the agent’s OpenAI-compatible provider at the same base URL. Tool calls use /v1/chat/completions; the server translates the model’s supported native grammar into structured tool calls.

An agent’s system instructions and tool definitions count towards the prompt length. Start with a short policy and a small tool set if the model repeats instructions or fails to choose tools. Parser support alone does not guarantee a model will use tools reliably. Check the released tool-use matrix for model-specific evidence.

Manage servers in your browser

Run:

krasis manager

Or open Krasis Manager from the Windows Start menu. The default dashboard is http://127.0.0.1:8090. It shows GPUs, selected models, settings and startup progress, and lets you validate, apply or stop a model server.

Manager is a separate control service. Its default local access boundary and owner token are separate from inference API keys. Keep it local unless you deliberately configure its LAN mode; the dashboard includes instructions for its management API.

Next: API and authentication.