A lightweight web-based tool to easily switch LLM providers and models for OpenClaw — no JSON editing required.
Home page: real-time display of the active model, one-click provider selection
After selecting a provider, Base URL and model list are automatically populated
Prerequisite: OpenClaw must be installed and initialized (~/.openclaw/openclaw.json must exist)
# Clone the repo
git clone https://github.com/yuanrengu/configadpter.git
cd configadpter
# Start the tool (zero dependencies — no pip install needed)
python3 openclaw_config.py
# Browser opens automatically at http://localhost:7890# Stop: press Ctrl+C in the terminal
# If running in background, kill the process with:
lsof -ti :7890 | xargs kill -9Requirements: Python 3.8+, standard library only — no
pip installneeded.
| Feature | Description |
|---|---|
| 📊 Live Status | Shows the active provider, model, and base URL at a glance |
| 🎛 Provider Switcher | Click a card to select a provider; Base URL and models auto-fill |
| 🔑 API Key Protection | Leave blank to keep the existing key — it will never be overwritten |
| 🗂 Model Management | Pick from built-in templates or add custom model IDs |
| 💾 Auto Backup | Creates a timestamped backup before every write: openclaw.json.bak.tool.YYYYMMDD_HHMMSS |
| 🔍 JSON Preview | Preview the exact changes before applying |
| ♻ Hot Reload Frontend | Edit index.html and refresh — no server restart needed |
| Provider | Base URL | Notes |
|---|---|---|
| 🌙 Moonshot (Kimi) | api.moonshot.cn |
Available in China, kimi-k2.5 / kimi-latest |
| 🔵 DeepSeek | api.deepseek.com |
Cost-effective; V3 + R1 reasoning |
| 🤖 OpenAI | api.openai.com |
GPT-4o, o1, o3, o3-mini |
| 🧠 Anthropic | api.anthropic.com |
Claude Opus 4.6 / Sonnet 4.6 (latest) |
| ✨ Google Gemini | generativelanguage.googleapis.com |
Gemini 2.5 Pro / Flash |
| 🔀 OpenRouter | openrouter.ai/api |
Multi-model aggregation proxy |
| 🚀 Groq | api.groq.com |
Ultra-fast inference (Llama, DeepSeek, etc.) |
| 🌊 SiliconFlow | api.siliconflow.cn |
Qwen3, DeepSeek, QwQ — CN-hosted |
| 🇨🇳 ZhipuAI | open.bigmodel.cn |
GLM-4 series, GLM Z1 reasoning |
| 🦙 Ollama | localhost:11434 |
Locally hosted offline models |
| ⚙️ Custom | Any URL | Any OpenAI-compatible API |
When you apply a model switch, the tool backs up the current config first, then only modifies these 4 areas in ~/.openclaw/openclaw.json. Everything else is left untouched.
models.providers.<provider> ← Rebuilds the target provider's connection config
auth.profiles.<provider>:default ← Adds or updates the auth profile entry
agents.defaults.model.primary ← Sets "provider/modelId" (the core field OpenClaw reads)
agents.defaults.models.<primary> ← Adds the model alias (display name)
API Key handling: If the key field is left blank, the existing key is inherited — it will never be lost.
channels.* ← Telegram / Discord channel settings
gateway.* ← Port, auth token, node config
tools.* ← Tool profiles
commands.* ← CLI settings
session.* ← Session config
hooks.* ← Internal hook config
wizard.* ← Wizard state
Other providers' configurations
The version shown in the top-right corner (e.g. v2026.3.2) comes from meta.lastTouchedVersion in openclaw.json. It is written and maintained by OpenClaw itself — this tool only reads and displays it.
Usually no. The OpenClaw Gateway monitors openclaw.json for changes and hot-reloads most configuration updates automatically.
Note: When switching to a brand-new provider that was never configured before, OpenClaw may need a moment to initialize the new connection. If the new model doesn't take effect immediately, check the OpenClaw web console or restart the OpenClaw process once.
- Click the 🔵 DeepSeek card
- Enter your DeepSeek API Key (
sk-...) - Select a model (e.g. DeepSeek Chat V3)
- Click ✅ Apply
- Click the 🌙 Moonshot (Kimi) card (shows "✓ Active")
- Leave the API Key blank (the existing key is reused automatically)
- Select a model and click Apply
- Click the ⚙️ Custom card
- Enter the Base URL (e.g.
https://your-api.com/v1) and API Key - Type a model ID and display name, then click Add
- Select the model and click Apply
.
├── openclaw_config.py # Backend: provider templates, HTTP server, config read/write
├── index.html # Frontend: Web UI (HTML / CSS / JS, dark theme)
├── docs/
│ ├── screenshot_home.png
│ └── screenshot_deepseek.png
└── README.md
OpenClaw paths:
- Config file:
~/.openclaw/openclaw.json - Backup dir:
~/.openclaw/openclaw.json.bak.tool.*
- The server only listens on
127.0.0.1:7890— never exposed to the network - Every write is preceded by an automatic backup, recoverable from
~/.openclaw/ - API keys are never pre-filled in the form and never printed to the terminal