Client setup · minirouter

Copy the two values

Every OpenAI-compatible client takes the same base URL and adds /chat/completions itself.

Base URLhttps://api.minirouter.sh/v1

API keyGet a key →Starts with mr-live-

Configure Zed

settings.json (zed: open settings file) → language_models

language_models → openai_compatibleminirouter
api_urlhttps://api.minirouter.sh/v1
API Keymr-live-… (your key)Use your full key. Placeholders will not authenticate.

settings.json

{
  "language_models": {
    "openai_compatible": {
      "minirouter": {
        "api_url": "https://api.minirouter.sh/v1",
        "available_models": [
          {
            "name": "zai/glm-5.3-flash",
            "display_name": "GLM 5.3 Flash (minirouter)",
            "max_tokens": 1000000
          },
          {
            "name": "openai/gpt-5.6-sol",
            "display_name": "GPT 5.6 Sol (minirouter)",
            "max_tokens": 1050000
          },
          {
            "name": "spacexai/grok-4.6",
            "display_name": "Grok 4.6 (minirouter)",
            "max_tokens": 500000
          }
        ]
      }
    }
  }
}

Full Zed guide →

Check the connection

Run this after saving. A reply confirms the base URL, key, and model together.

Terminal

# bash / zsh: enter the full key at the hidden prompt
printf 'MiniRouter API key: '
read -rs MINIROUTER_KEY
export MINIROUTER_KEY
printf '\n'
curl https://api.minirouter.sh/v1/chat/completions \
  -H "Authorization: Bearer $MINIROUTER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"meta/llama-3.3-70b","messages":[{"role":"user","content":"ping"}]}'

SDK examples

Calling from code instead? The base URL stays the same.

Cohere SDK

import os
import cohere

client = cohere.ClientV2(
    api_key=os.environ["MINIROUTER_KEY"],
    base_url="https://api.minirouter.sh",
)

result = client.rerank(
    model="cohere/rerank-v3.5",
    query="Which document is about London?",
    documents=["London is in the UK.", "Paris is in France."],
)
print(result.results)

Models and errors

Model IDs are author/name from the model catalog. Failed requests are explained in the error reference.