openclaw.md

March 31, 2026 · View on GitHub

OpenClaw with model routing (gpt4free)

OpenClaw is a self-hosted gateway that connects your favorite chat apps — WhatsApp, Telegram, Discord, iMessage, and more — to AI coding agents like Pi. You run a single gateway process on your own machine (or a server), and it becomes the bridge between your messaging apps and an always-available AI assistant.

Who is it for? Developers and power users who want a personal AI assistant they can message from anywhere — without giving up control of their data or relying on a hosted service.

This guide shows how to define an openclaw named model route in gpt4free using config.yaml, set provider API keys in .env, and verify behavior.

1. Why openclaw?

2. Add route to config.yaml

Edit your cookies/config.yaml (or ~/.config/g4f/cookies/config.yaml):

models:
  - name: "openclaw"
    providers:
      - provider: "GeminiCLI"
        model: "gemini-3-flash-preview"
        condition: "quota.models.gemini-3-flash-preview.remaining > 0 and error_count < 3"
      - provider: "Antigravity"
        model: "gemini-3-flash"
        condition: "quota.models.gemini-3-flash.quotaInfo.remainingFraction > 0 and error_count < 3"
      - provider: "PollinationsAI"
        model: "openai"
        condition: "balance > 0 or error_count < 3"

Notes

  • condition is optional; if not provided the provider is always eligible.
  • Missing quota fields are treated as 0.0.
  • error_count tracks recent failures for each provider (last hour).

3. Set API keys in .env

In your project root or cookies folder:

GEMINI_API_KEY=your_gemini_key
ANTIGRAVITY_API_KEY=your_antigravity_key
POLLINATIONS_API_KEY=your_pollinations_key
OPENAI_API_KEY=your_openai_key

Providers may require additional environment variables; consult individual provider docs in g4f/Provider/.

4. Use the route in code

from g4f.client import Client
client = Client()

res = client.chat.completions.create(
    model="openclaw",
    messages=[{"role": "user", "content": "Hello from OpenClaw"}],
)
print(res.choices[0].message.content)

Or CLI:

g4f client "Hello" --model openclaw --debug

5. Debug and verify

  • --debug logs show selected provider and route decisions.
  • openclaw will try each provider in order and fallback when conditions fail.
  • Use g4f client --model openclaw to validate end-to-end.

6. Advanced

  • Add other providers (OpenAI, Yupp, etc.) under the same openclaw route.
  • Use condition: "error_count < 3" for generic resiliency.
  • If you need special provider settings, set them in g4f/config.py or provider-specific environment variables.

Quick startup (Pollinations-style)

  1. Get API key from pollinations.ai (or another provider).
  2. Set it in .env as above.
  3. Choose your run mode and model in config.yaml:
models:
  - name: "openclaw"
    providers:
      - provider: "GeminiCLI"
        model: "gemini-3-flash-preview"
      - provider: "Antigravity"
        model: "gemini-3-flash"
      - provider: "PollinationsAI"
        model: "openai"
  1. Call with model="openclaw" and fallback happens automatically.

Optional tooling

  • add web search by configuring g4f provider flags (e.g. web_search=True where supported)
  • add vision tools via provider-specific params or additional providers
  • for script driven setup, run an install script that writes config.yaml from a template

gpt4free setup script

Use the included script:

curl -fsSL https://raw.githubusercontent.com/xtekky/gpt4free/main/scripts/setup-openclaw.sh | bash -s -- <POLLINATIONS_API_KEY>

OpenClaw builder flow

  • API + key retrieval: pollinations.ai account, free credits
  • model router: openclaw route
  • custom model from gpt4free is a provider chain instead of a service URL

Return to Documentation