Configure web search tools like Brave using openclaw configure --section web, which stores values such as tools.web.search.apiKey in your JSON5 config file. Enables autonomous web search.
Upgrade by running npm i -g openclaw@latest or use the install script with --no-onboard flag. It detects existing installs and upgrades in place. Treat updates like infrastructure changes with rollback ready.
Create multiple agents with openclaw agents add, each with its own model and workspace. Bind Discord channels or roles to specific agent IDs for multi-agent routing.
Start with minimal access and add capabilities one-by-one. Run openclaw security audit --fix to apply safe guardrails. Use dmPolicy pairing or allowlists, and enable mention gating.
When running OpenClaw on a VPS, install the gateway as a daemon so it persists after SSH disconnection. Place environment variables in ~/.openclaw/.env since the gateway service doesn't inherit SSH shell variables.
If the browser control service isn't responding, restart the gateway via the menubar or terminal. For Chrome extension relay issues, ensure Chrome is open and click the OpenClaw extension icon to re-attach.
On Windows, the openclaw folder may be locked by a running process during updates. Stop the gateway first with openclaw gateway stop, close all processes, then run the update again.
Telegram offers a cleaner personal assistant UX with 1:1 chat and SMS-style push notifications. Discord is better for community collaboration with moderation and role features.
Configure OpenRouter by setting the OPENROUTER_API_KEY env var, then select an OpenRouter model ID with the openrouter/ prefix. Note that openrouter/auto routing is not supported.
Each OpenClaw agent can have its own isolated memory and workspace. Configure per-agent memory isolation in the agents section of openclaw.json to prevent agents from sharing context.
OpenClaw memory has two layers: source-of-truth Markdown files in the agent workspace, and a retrieval/index backend for search. Data must be written to disk to persist across restarts.
Run OpenClaw in Docker with Node 20-slim, install via npm, expose port 18789, and use volume mounts for persistent config. Access the Docker instance from other machines on the same network.
The recommended way to install OpenClaw on Windows is via WSL2 (Ubuntu). Run the install script inside WSL2, then run the onboarding wizard. Native Windows is not fully supported.
For a 4090 GPU running local models with OpenClaw, models like GLM 4.7 flash work well. Keep context windows at 16k-32k tokens for the best balance of speed and stability.
OpenClaw doesn't use the Claude Desktop MCP config directly. Use the bundled mcporter skill as an MCP client/manager, or use openclaw-mcp-bridge to aggregate multiple MCP servers.
Set up Ollama by configuring the API key via openclaw config set models.providers.ollama.apiKey and select your model with the ollama/ prefix. Works with any Ollama-compatible endpoint.
Change your default model using openclaw models set or by editing agents.defaults.model.primary in your openclaw.json config. Restart the gateway after making changes.
The Self-Improving Agent skill lets OpenClaw learn from interactions and improve over time. Install it from ClawHub and configure it to write improvements to the agent's workspace.
Create custom skills by writing a SKILL.md file that defines the skill's behavior and placing it in the correct skills directory. OpenClaw loads skills from specific locations on startup.
If OpenClaw ignores its loaded skills even when prompted, verify skills are properly installed and listed. Check with openclaw skills list or ask the agent directly.
If the gateway runtime stops unexpectedly, use openclaw doctor --fix to automatically diagnose and restart, or manually restart with openclaw gateway restart.
When the gateway won't connect, run openclaw gateway status on the same machine to see the last error. Ensure you have a valid gateway token and correct network configuration.
If the gateway closes during initial setup with a 'Health check failed: gateway closed (1006 abnormal closure)' error, check that no other process is using port 18789 and re-run the setup command.
Connect ACP as a remote node to your OpenClaw gateway by configuring the ACP bridge mode settings. This allows you to use remote compute resources while keeping your gateway local.
If your OpenClaw gateway gets disconnected and the terminal shows 'command not sent gateway disconnected,' restart the gateway using the menubar app or run openclaw gateway restart in a new terminal session.
Strategies for managing token limits when pairing Claude Code with OpenClaw. Includes tips on switching providers, using local models as fallback, and optimizing token usage.
配置浏览器模式：使用 openclaw config set gateway.nodes.browser.mode auto，然后重启Gateway。解决浏览器工具无法调用的常见问题。
Run OpenClaw on a server without a window manager or graphical interface. Configure the gateway for remote access and manage it entirely via CLI.
Gateway无法启动或意外关闭？运行 openclaw doctor --fix 自动重启，或使用 openclaw gateway restart 手动重启。涵盖常见的Gateway连接问题。
Access the OpenClaw dashboard to retrieve your gateway token. It opens in your browser with the token on the same machine, or shows the URL for remote access. Default port is 18789.
Tool-calling reliability is the top priority. The recommended approach: run OpenClaw locally while using a cloud API model for the main agent, keeping runtime and logs on your machine.
通过npm安装OpenClaw，初始化并运行Gateway，使用ClavHub搜索和安装技能。包含常见安装问题的排查方法。
OpenClaw can connect to local models via LM Studio's OpenAI-compatible HTTP server. On Windows, run the Gateway inside WSL2 and point it at Ollama for a fully local setup.
OpenClaw supports multiple web search providers including Brave Search and Perplexity Sonar. Learn how to configure your preferred search provider for agent web access.
Step-by-step guide covering initial installation, gateway onboarding, and security best practices. Keep the gateway private by default with loopback bind and token authentication.
No tips found matching your search.