MegaBrain University · guide

OpenClaw Custom Provider Setup: Claude, GPT-6, Gemini and Kimi on One API Key

Updated November 11, 2026 · 15 min setup

What is OpenClaw

OpenClaw (formerly Clawdbot and Moltbot) is an open-source personal AI assistant. A gateway process runs on your computer or a server, connects to the chat apps you already use — WhatsApp, Telegram, Slack, Discord, Microsoft Teams, iMessage, Google Chat, Signal and more — and acts with tools, skills from the ClawHub registry and scheduled jobs.

OpenClaw needs a model provider. Its config accepts any OpenAI-compatible endpoint as a custom provider, which is how you connect MegaBrain: one key and balance for Claude, GPT, Gemini, Grok and open models instead of a separate account per lab.

Setup, step by step

1. Step 1. Get a MegaBrain API key

   Sign up with GitHub, Google or email, add funds on the Balance page and create a key on the API key page. The key works for every text, image and video model, and each request is paid from your balance — no subscription.

   Put the key in an environment variable so it never lands in a config file you share:

   # ~/.zshrc or ~/.bashrc
   export MEGABRAIN_API_KEY="your-megabrain-key"

   Copy

   curl https://getmegabrain.com/api/gateway/v1/chat/completions \
     -H "Authorization: Bearer $MEGABRAIN_API_KEY" \
     -H "Content-Type: application/json" \
     -d '{"model": "anthropic/claude-sonnet-5.5", "messages": [{"role": "user", "content": "Say OK"}]}'

   Copy

   Check the key before you touch the agent. A JSON reply with choices means the key, the balance and the model all work:

2. Step 2. Install OpenClaw

   On macOS, Linux or WSL2 (the script starts onboarding when it finishes):

   curl -fsSL https://openclaw.ai/install.sh | bash

   Copy

   With npm: npm install -g openclaw@latest --allow-scripts=openclaw, then openclaw onboard --install-daemon. On Windows: iwr -useb https://openclaw.ai/install.ps1 | iex.

3. Step 3. Add MegaBrain as a custom provider

   The fastest way is onboarding with a custom API key. Non-interactive setup requires --accept-risk, your acknowledgement that the agent gets full access to the machine. Without --custom-model-id, OpenClaw reads /models and saves every model it finds:

   openclaw onboard --non-interactive --accept-risk \
     --auth-choice custom-api-key \
     --custom-base-url "https://getmegabrain.com/api/gateway/v1" \
     --custom-model-id "anthropic/claude-sonnet-5.5" \
     --custom-api-key "$MEGABRAIN_API_KEY" \
     --custom-compatibility openai

   Copy

   Or write the provider into ~/.openclaw/openclaw.json yourself (OpenClaw reloads it on save):

   {
     agents: { defaults: { model: { primary: "megabrain/anthropic/claude-sonnet-5.5" } } },
     models: {
       mode: "merge",
       providers: {
         megabrain: {
           baseUrl: "https://getmegabrain.com/api/gateway/v1",
           apiKey: "${MEGABRAIN_API_KEY}",
           api: "openai-completions",
           models: [
             { id: "anthropic/claude-sonnet-5.5", name: "Claude Sonnet 5.5",
               input: ["text", "image"], contextWindow: 1000000, maxTokens: 32000 },
             { id: "openai/gpt-6.1-sol", name: "GPT-6.1 Sol",
               input: ["text", "image"], contextWindow: 1050000, maxTokens: 32000 },
             { id: "google/gemini-3.8-flash", name: "Gemini 3.8 Flash",
               input: ["text", "image"], contextWindow: 1048576, maxTokens: 32000 },
             { id: "mb-auto/free", name: "Auto Free",
               input: ["text"], contextWindow: 128000, maxTokens: 8000 }
           ]
         }
       }
     }
   }

   Copy

   A model reference splits at the first slash: in megabrain/anthropic/claude-sonnet-5.5 the provider is megabrain and the model is anthropic/claude-sonnet-5.5. Each model you reference must be listed in models. The config is validated strictly, so a misspelled key stops the gateway — check the log.

4. Step 4. Connect a channel and test

   Open the Control UI at http://127.0.0.1:18789 or run onboarding again to connect WhatsApp, Telegram or Slack. Send the assistant a message; the reply comes from the model in primary.

Which model to choose

OpenClaw uses tools on almost every turn, so pick a model with reliable tool calling. Claude Sonnet 5.5 is the default; put a cheaper model on high-volume channels. Set input: ["text", "image"] on models that see images so photos sent in chat reach them.

Every model with its price and context window is in the catalog. They all work with the same key.

What you can do

Business use cases

You answer customers of our studio in WhatsApp. Use only facts from ~/business/faq.md. If someone wants to book, collect name, date and service and append a row to ~/business/bookings.csv. If you are unsure, say a manager will reply within an hour and notify me in Telegram.

Troubleshooting

FAQ