Give Your Agent a Brain, Voice & Skills
Your agent is installed but can't think, talk, or act yet. Three things to do: ① Connect an LLM API as its brain, ② Set up a messaging tool as its voice, ③ Install Skills as its hands. Once fed, your agent comes alive.
Deployed via cloud platform?
If you deployed via Tencent Cloud, Alibaba Cloud, etc., model and messaging config is usually done during deployment. Check your cloud platform tutorial:
Step 1: Pick an LLM as the Brain
OpenClaw supports all major LLMs. Chinese users: start with DeepSeek or GLM-5. Global users: Claude or GPT
Not sure which one?
Start with DeepSeek — cheapest, great for Chinese, free credits, easiest setup. Get it running first, you can switch anytime.
DeepSeek
Rising star in Chinese LLMs, extremely affordable with great Chinese performance. Free credits for new users
Step-by-step
- 1Visit platform.deepseek.com, click 'Sign Up' in the top right
- 2Register with your phone number and verify via SMS
- 3Go to Console, select 'API Keys' from the left menu
- 4Click 'Create API Key', name it (e.g. 'My OpenClaw')
- 5Copy the generated Key (format: sk-xxxxxxxx) — shown only once
- 6Open .env in OpenClaw and enter the config below
.env config
LLM_PROVIDER=deepseek LLM_API_KEY=sk-your-deepseek-key LLM_MODEL=deepseek-chat
GLM-5(智谱)
Zhipu AI's flagship model, exceptional at agent planning and tool use, top-rated in the community
Step-by-step
- 1Visit open.bigmodel.cn, click 'Get Started' to register
- 2Complete identity verification (ID card required, ~1-2 min)
- 3After verification, go to Console → 'API Keys' in the left menu
- 4Click 'Create API Key' and copy the generated key
- 5Note: API base URL is open.bigmodel.cn/api/paas/v4/
- 6Open .env in OpenClaw and enter the config below
.env config
LLM_PROVIDER=zhipu LLM_API_KEY=your-glm-api-key LLM_MODEL=glm-5 LLM_API_BASE=https://open.bigmodel.cn/api/paas/v4/
MiniMax 2.5
MiniMax's latest model, fast response with OpenAI-compatible API format, stable for general agent tasks
Step-by-step
- 1Visit platform.minimaxi.com, register an account
- 2Complete identity verification
- 3Go to Console, find 'API Key Management'
- 4Create a new API Key and copy it
- 5Confirm API base URL: api.minimaxi.com/v1
- 6Open .env in OpenClaw and enter the config below
.env config
LLM_PROVIDER=minimax LLM_API_KEY=your-minimax-key LLM_MODEL=minimax-2.5 LLM_API_BASE=https://api.minimaxi.com/v1
Qwen-Max(通义千问)
Alibaba Cloud's LLM, strong Chinese comprehension, ideal for enterprise with deep Alibaba ecosystem integration
Step-by-step
- 1Visit dashscope.console.aliyun.com, log in with your Alibaba Cloud account
- 2Enable 'DashScope' model service and accept the terms
- 3Go to Console → 'API Key Management'
- 4Click 'Create New API Key', copy the key (sk-xxxxxxxx format)
- 5Open .env in OpenClaw and enter the config below
.env config
LLM_PROVIDER=openai LLM_API_KEY=sk-your-qwen-key LLM_MODEL=qwen-max LLM_API_BASE=https://dashscope.aliyuncs.com/compatible-mode/v1
Claude Opus 4.6 / Sonnet 4.6
Anthropic's latest models, industry-leading code and complex reasoning. Opus 4.6 for flagship, Sonnet 4.6 for value
Step-by-step
- 1Visit console.anthropic.com, click 'Sign Up'
- 2Verify your email and phone number
- 3Go to Console → 'API Keys' in the left menu
- 4Click 'Create Key' and name it
- 5Copy the key (format: sk-ant-api03-xxxxxxxx) — shown only once
- 6Open .env in OpenClaw and enter the config below
.env config
LLM_PROVIDER=anthropic LLM_API_KEY=sk-ant-api03-your-key LLM_MODEL=claude-opus-4-6 # 或性价比之选:LLM_MODEL=claude-sonnet-4-6
GPT-5.2
OpenAI's latest model, strong multimodal capabilities, most mature global developer ecosystem
Step-by-step
- 1Visit platform.openai.com, sign up or log in
- 2Navigate to the 'API keys' page
- 3Click 'Create new secret key' and name it
- 4Copy the key (format: sk-proj-xxxxxxxx) — shown only once
- 5Open .env in OpenClaw and enter the config below
.env config
LLM_PROVIDER=openai LLM_API_KEY=sk-proj-your-openai-key LLM_MODEL=gpt-5.2
Verify Model Connection
After configuring .env, run the test command to verify:
cd openclaw python -m openclaw.test_connection
Success looks like:
✅ Connected! Model: deepseek-chat Response: "Hello! I'm DeepSeek, happy to help." Latency: 1.2s Tokens: input 12 / output 18
Step 2: Connect a Messaging Tool as the Voice
Pick a messaging tool you already use daily, so you can chat with your agent right from your messenger
Not sure which one?
For Chinese users: Feishu — easy setup, great docs, free for personal use. Global users: Telegram — just one token, 30 seconds to set up.
Feishu / Lark
ByteDance's workspace platform with robust bot capabilities. Easy setup with rich community tutorials
Step-by-step
- 1Visit open.feishu.cn, log in with your Feishu account
- 2Click 'Create Custom App', enter app name (e.g. 'My OpenClaw')
- 3In app details, find 'Credentials', copy App ID and App Secret
- 4Select 'Bot' in left menu, enable bot capability
- 5In 'Event Subscriptions', set callback URL to your OpenClaw address
- 6Get the Verification Token from the event subscription page
- 7Open .env in OpenClaw and enter the config below
- 8Publish the app version, search for your bot in Feishu to test
.env config
# 飞书机器人 FEISHU_APP_ID=cli_your_app_id FEISHU_APP_SECRET=your_app_secret FEISHU_VERIFICATION_TOKEN=your_verification_token FEISHU_ENCRYPT_KEY=your_encrypt_key # 可选,开启加密时填写
WeCom
Tencent's enterprise communication platform, ideal for teams already using WeCom
Step-by-step
- 1Visit work.weixin.qq.com, log in with admin account
- 2Go to 'App Management' → 'Custom', click 'Create App'
- 3Fill in app name and logo, select visibility scope
- 4After creation, find AgentID and Secret in app details
- 5Find your CorpID on the 'My Enterprise' page
- 6Configure 'Receive Messages' callback URL in app settings
- 7Set Token and EncodingAESKey (auto-generated, save them)
- 8Open .env in OpenClaw and enter the config below
.env config
# 企业微信 WECOM_CORP_ID=your_corp_id WECOM_AGENT_ID=your_agent_id WECOM_SECRET=your_secret WECOM_TOKEN=your_callback_token WECOM_ENCODING_AES_KEY=your_encoding_aes_key
DingTalk
Alibaba's workspace platform, supports group bots and direct chat for teams using DingTalk
Step-by-step
- 1Visit open.dingtalk.com, log in by scanning QR code with DingTalk
- 2Click 'Create App', select 'Enterprise Internal' → 'Bot'
- 3Fill in app name and description, complete creation
- 4In app details, find AppKey and AppSecret
- 5Go to 'Bot & Message Push', enable bot configuration
- 6Set message receiving URL to your OpenClaw callback address
- 7In 'Permission Management', add required permissions
- 8Open .env in OpenClaw and enter the config below
.env config
# 钉钉机器人 DINGTALK_APP_KEY=your_app_key DINGTALK_APP_SECRET=your_app_secret
WeChat (Personal)
Connect via bridge tools like wcferry. More complex setup with some risk, recommended for experienced users
Step-by-step
- 1Prepare a secondary WeChat account for testing (risk of ban, don't use main account)
- 2Install wcferry bridge tool (requires Windows environment or VM)
- 3Download the matching WeChat client version
- 4Start wcferry service and log in to WeChat
- 5Configure wcferry connection URL in OpenClaw
- 6Open .env and enter the config below
- 7Send a test message to verify it works
.env config
# 微信接入(需配合 wcferry 桥接工具) WECHAT_ENABLED=true WECHAT_WCFERRY_API=http://localhost:10086
Telegram
One of the world's most popular messengers. Mature Bot API, simplest setup — just one token needed
Step-by-step
- 1Search for @BotFather in Telegram and start a chat
- 2Send /newbot, follow prompts to set bot name and username
- 3BotFather returns a Bot Token (format: 123456789:ABCdefGHI...)
- 4Copy this Token and save it securely
- 5Open .env in OpenClaw and enter the config below
- 6Restart OpenClaw, find your bot in Telegram to test
.env config
# Telegram Bot TELEGRAM_BOT_TOKEN=123456789:ABCdefGHIjklMNOpqrsTUVwxyz
Step 3: Install Skills as Hands & Feet
Skills are your agent's capability plugins — web search, file operations, messaging, automation. Without Skills, your agent can only chat
Recommended starter Skills
Start with these essentials: web search, file-manager, code-runner. These cover 80% of daily use cases.
How to Install Skills
Install Skills via command line with a single command, or use the one-click install in the OpenClaw management panel.
Command line install example
Install search skill
openclaw skill install @openclaw/search
Install file manager skill
openclaw skill install @openclaw/file-manager
List installed skills
openclaw skill list
Want more? Browse all 5400+ available skills in the skill store, filter by category, and install with one click:
Browse all skills →All Set! See How Others Use It
Your agent now has a brain, voice, and skills — it can think, talk, and act. Check out real use cases to see what others are building: social media automation, email handling, code review... find your own style.