Connect your tools

Most tools need three things: a base URL, your API key and a model id. This page gives them for the common ones. Start by creating a key, see authentication, and by picking a model id from models.

Claude Code#

export ANTHROPIC_BASE_URL="https://ai.shvia.org"
export ANTHROPIC_AUTH_TOKEN="your key"
export ANTHROPIC_MODEL="a model id from GET /v1/models"
claude

To make it permanent, add these lines to your shell profile. Claude Code then talks to the gateway through /v1/messages, and the model you picked answers, which does not have to be an Anthropic model.

Claude Code sends settings that the gateway ignores, among them max_tokens and thinking, and images are dropped. Tool calls are translated between the two dialects. A single session can use many requests, so watch the request limit.

OpenAI SDKs and tools that use environment variables#

export OPENAI_BASE_URL="https://ai.shvia.org/v1"
export OPENAI_API_KEY="your key"

Any program that reads those two variables will use the gateway, and you choose the model in the program. In code:

import os
from openai import OpenAI

client = OpenAI(base_url="https://ai.shvia.org/v1", api_key=os.environ["SHVIA_API_KEY"])

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://ai.shvia.org/v1", apiKey: process.env.SHVIA_API_KEY });

client.chat.completions.create and client.models.list work, and so does client.responses.create for the stateless cases described in responses. client.models.retrieve, client.embeddings.create, client.completions.create and the image endpoints have no counterpart.

Anthropic SDKs#

import os
import anthropic

client = anthropic.Anthropic(base_url="https://ai.shvia.org", api_key=os.environ["SHVIA_API_KEY"])

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({ baseURL: "https://ai.shvia.org", apiKey: process.env.SHVIA_API_KEY });

Use a model id from the gateway in model. See messages for what is ignored.

Cline and Roo Code#

In the extension's settings, choose these values:

These tools talk to chat completions, which accepts only string content. If an extension reports a 422 that says a message content must be a string, it is sending content parts. Images are not supported on that endpoint.

Continue#

In Continue's configuration, add a model with the openai provider and the gateway as its API base:

{
  "models": [
    {
      "title": "ShvIA",
      "provider": "openai",
      "model": "a model id from GET /v1/models",
      "apiBase": "https://ai.shvia.org/v1",
      "apiKey": "your key"
    }
  ]
}

Newer versions of Continue read a YAML file instead of JSON. The fields are the same: provider: openai, model, apiBase and apiKey. Where the file lives depends on your version.

curl#

curl https://ai.shvia.org/v1/chat/completions \
  -H "Authorization: Bearer $SHVIA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "a model id from GET /v1/models", "messages": [{"role": "user", "content": "hi"}]}'

Let the app write the configuration#

The account page of the ShvIA web app has a "Connect my CLI" section. It lists the models of GET /v1/models, and prints the settings for Claude Code, Cline, Roo Code, Continue, the environment variables and curl, with a placeholder where the key goes. The desktop app can also save the settings on your computer for you, in the configuration file of the tool.

When something does not work#