OpenClaw Configuration Directory
OpenClaw (also known as Clawdbot) centralizes all of its configuration, state data, workspaces, and skills in the core directory ~/.openclaw/ (Linux/macOS) or %USERPROFILE%\.openclaw\ (Windows) under the user's home directory.
~/.openclaw/is the root configuration directory of the entire system; on first installation/runopenclaw onboardoropenclaw onboard --install-daemonit is automatically created.
~/.openclaw/ ├── openclaw.json # 主配置文件(JSON/JSON5) ├── workspace/ # 你的 AI “灵魂”文件夹(推荐 git 版本控制) │ ├── SOUL.md # 人格设定(语气、风格) │ ├── USER.md # 你的个人信息(让 AI 更懂你) │ ├── MEMORY.md # 长期记忆(手动可编辑) │ ├── IDENTITY.md # Agent 名称、形象 │ ├── AGENTS.md # 多 Agent 路由规则 │ ├── BOOT.md # 启动提示词 │ ├── HEARTBEAT.md # 每日检查清单 │ └── skills/ # 已安装技能(每个技能一个子文件夹) ├── agents/<cid>/ # 每个 Agent 的独立状态 ├── memory/<cid>.sqlite # 向量记忆库 ├── credentials/ # API Key、OAuth(旧版) ├── skills/ # 全局技能包 └── secrets.json # 加密凭证(可选)
Most Important Files:
openclaw.json: Global settings (model, channels, ports, security policies)workspace/all under.mdFiles:You can directly edit with VS Code and it takes effect in real time.!
Command to view/modify configuration:
openclaw configure # 交互式配置向导 openclaw config file # 显示完整路径 openclaw config get agent.model openclaw config set agent.model "anthropic/claude-3-5-sonnet" openclaw config validate # 校验合法性
| Path | Purpose Description |
|---|---|
~/.openclaw/openclaw.json |
Main Configuration File(Most Important!) Stores global settings (model, Gateway mode, bind address, Agents default values, plugins, etc.). Can be viewed/modified with the command: • openclaw config file(displays the full path)• openclaw config get <json路径>• openclaw config set <json路径> <值> |
~/.openclaw/workspace/ |
Default Workspace(The location of the Agent's "soul") Stores the following Markdown files (can be edited directly and take effect): • SOUL.md: AI personality/tone settings• USER.md: Your personal preferences and background• MEMORY.md: Long-term memory records• AGENTS.md: Instruction descriptions• IDENTITY.md: Name/Theme• BOOT.md: Startup configuration• HEARTBEAT.md: Periodic checklist |
~/.openclaw/agents/<cid>/ |
State directory of a single agent (<cid>for instance ID) |
~/.openclaw/agents/<agentId>/agent/auth-profiles.json |
API Key & OAuth credentials(recommended location in new version) Legacy version may be in ~/.openclaw/credentials/ |
~/.openclaw/memory/<cid>.sqlite |
Vector index storage (for memory search) |
~/.openclaw/skills/ |
Global skill directory Via openclaw skills installorclawhub installAll installed skill packages are placed here (each skill is a subfolder containing SKILL.md) |
~/.openclaw/memory/ |
Long-term memory related files (SQLite + vector embeddings) |
/tmp/openclaw/*.log |
Gateway service logs (for debugging) |
1. Main Configuration File
Path: ~/.openclaw/openclaw.json
Description:OpenClaw's core configuration file, containing all system-level settings.
Main configuration items:
- gateway- Gateway service configuration
mode: Gateway mode (e.g., "local")port: Gateway port (default 18789)bind: Bind address (default 127.0.0.1)token: Access token, used for Web UI authentication
- models- AI model configuration
- Default model settings
- API authentication information
- messages- Message processing configuration
- TTS (text-to-speech) settings
- Message format configuration
Example configuration snippet:
{
"gateway": {
"mode": "local",
"port": 18789,
"bind": "127.0.0.1",
"token": "c9917c5a066beeb26266d09baed99495e7563b33c771e89a"
}
}
Operation suggestions:
- After first installation, it is
openclaw onboardautomatically generated by the wizard - After modifying the configuration, restart the Gateway service for it to take effect.
- The token value is used for Web UI access; please keep it safe.
2. Workspace Directory
Path: ~/.openclaw/workspace/
Description:OpenClaw's default working directory, where all AI-generated files, temporary files, and output files from user requests are saved.
Main uses:
- Default location for AI agent file read/write
- Code execution output storage
- Document generation and editing
- Temporary data storage
Permission notes:
- By default, AI can only access this directory and its subdirectories.
- If access to other paths is required, configure
filesystem-mcpskills and grant additional permissions - Windows users may need to copy files from the workspace to other locations.
Usage suggestions:
- Regularly clean up unneeded temporary files
- It is recommended to back up important outputs outside the workspace.
- You can create project subdirectories here for organization and management.
3. Agent State Directory
Path: ~/.openclaw/agents/<cid>/
Description:Directory for storing agent state and configuration for each conversation.<cid>It is the conversation ID, and each conversation has independent configuration space.
Directory structure:
~/.openclaw/agents/<cid>/ ├── agent/ │ ├── auth-profiles.json # 该会话的 OAuth 和 API 密钥 │ └── ... └── ...
auth-profiles.json description:
- Stores the API authentication information used by this agent
- Contains OAuth tokens for various services
- The new version uses this path; the old version is stored in
~/.openclaw/credentials/
Data isolation:
- Authentication information for each conversation is independent
- Different API keys can be configured for different conversations
- Supports parallel operation of multiple agents
4. Authentication Credentials Directory (Legacy)
Path: ~/.openclaw/credentials/
Description:Credential storage location for the old version of OpenClaw.
Migration notes:
- The new version has migrated to
~/.openclaw/agents/<agentId>/agent/auth-profiles.json - After upgrading from the old version, credentials may need to be reconfigured
- It is recommended to use
openclaw models auth setupto reset
5. Memory Storage Directory
Path: ~/.openclaw/memory/
Description:Storage location for OpenClaw's persistent memory system, including vector indexes and conversation history.
Main files:
-
<cid>.sqlite- Vector index database- Stores semantic vectors of conversations
- Supports semantic search functionality
- Used for long-term memory retrieval
-
YYYY-MM-DD.md- Daily conversation logs- Markdown files named by date
- Records all conversation content for the day
- Facilitates manual review and archiving
Memory features:
- Vector search:Uses
memory search "关键词"to search historical conversations - Context association:Supports cross-conversation context understanding
- Long-term memory:The agent will 'remember' user preferences and habits
Maintenance recommendations:
- Regularly back up the memory directory
- Large databases may affect performance; consider regular archiving
- Log files can be used for auditing and troubleshooting
6. Skills Directory
Path: ~/.openclaw/skills/
Description:Installation location for globally shared skills. Skills are plugins that extend OpenClaw's functionality.
Skill management:
- Install skills:
npx clawhub install <skill-name> - List skills:
openclaw skills list - Search skills:
npx clawhub search <关键词>
Common skill examples:
filesystem-mcp- File system operationsgithub- GitHub integrationnano-pdf- PDF editingnotion/obsidian- Note synchronizationweather- Weather querysummarize- Content summary generation
Skills ecosystem:
- Community skills library ClawHub provides 500+ skills
- Supports custom skill development
- Skills can access external APIs and system resources
Configuration instructions:
- Each skill may require separate API Key configuration
- Some skills depend on system tools (e.g., GitHub CLI)
- macOS-exclusive skills are not available on Windows/Linux
7. Gateway Log Directory
Path: /tmp/openclaw/*.log
Description:Runtime logs of the gateway service, stored in the system temporary directory.
Log types:
- Gateway start/stop logs
- API request/response logs
- Error and exception logs
- Performance monitoring information
Log management:
- Logs in the temporary directory may be cleared after system restart
- Use
openclaw gateway --verboseDetailed logs can be viewed - When troubleshooting, check the log files first
Gateway service management:
openclaw gateway start # 启动网关 openclaw gateway status # 查看状态 openclaw gateway stop # 停止网关
Other Important Files
USER.md
Path:Usually located in the workspace or user-defined location
Description:User information file, containing personalized information about the user, helping AI better understand user needs.
Content examples:
- User preference settings
- Common workflows
- Project background information
- Custom instructions
Best Practices for Configuration File Management
1. Backup Strategy
Critical directory backup:
# 备份整个配置目录 tar -czf openclaw-backup-$(date +%Y%m%d).tar.gz ~/.openclaw/ # 仅备份核心配置 cp ~/.openclaw/openclaw.json ~/backups/ cp -r ~/.openclaw/memory/ ~/backups/memory/
Recommended backup frequency:
- Main configuration file: after each modification
- Memory database: once a week
- Skill configuration: after installing new skills
2. Security Recommendations
Sensitive information protection:
openclaw.jsonContains gateway token, should not be made publicauth-profiles.jsonContains API keys, needs encryption protection- Regularly change API keys
- Do not commit configuration files to version control systems
Permission settings:
# 确保配置目录仅用户可访问 chmod 700 ~/.openclaw/ chmod 600 ~/.openclaw/openclaw.json chmod 600 ~/.openclaw/agents/*/agent/auth-profiles.json
3. Migration and Upgrade
Version upgrade considerations:
- Before upgrading, back up the entire
~/.openclaw/directory - Check release notes for configuration changes
- Old version credentials may need migration
Cross-device sync:
- Can sync configuration files for multi-device consistency
- Note path differences (Windows vs Unix)
- Sensitive information is recommended to be configured separately
4. Troubleshooting
Configuration issue diagnosis:
# 检查配置文件语法 cat ~/.openclaw/openclaw.json | json_pp # 深度检查 openclaw doctor --deep # 查看网关状态 openclaw gateway status
Common issues:
- Authentication expired:Run
openclaw models auth setupReconfigure - Gateway fails to start:Check port occupancy, view log files
- Skills unavailable:Confirm skill dependencies are installed, check API keys
Environment Variable Configuration
OpenClaw supports configuration via environment variables:
# API 密钥环境变量 export ANTHROPIC_API_KEY="your-key-here" export OPENAI_API_KEY="your-key-here" export DEEPSEEK_API_KEY="your-key-here" # 自定义配置目录 export OPENCLAW_HOME="~/custom-path/.openclaw"
Environment variable priority:
- API Keys in environment variables are automatically detected by the onboard wizard
- Can override default values in configuration files
- Suitable for CI/CD environments and containerized deployment
Network and Deployment Configuration
Local Deployment
Default configuration:
- Bind address: 127.0.0.1 (local access only)
- Gateway port: 18789
Access method:
http://127.0.0.1:18789/#token=<your-token>
Cloud Deployment
OpenClaw supports deployment on cloud servers:
Supported platforms:
- Alibaba Cloud:https://www.aliyun.com/activity/ecs/clawdbot
- Tencent Cloud:https://cloud.tencent.com/developer/article/2624973
- 1Panel
- Docker container
Cloud configuration features:
- Need to configure public internet access permissions
- May need to configure firewall rules
- Supports domain binding and SSL certificates
Integration Platform Configuration
OpenClaw supports multiple instant messaging platforms:
Supported Platforms
- International: WhatsApp, Telegram, Discord, Slack, iMessage
- Domestic:Feishu, DingTalk
Configuration Location
Chat channel configuration is stored inopenclaw.json, including:
- Channel type (Slack/Feishu, etc.)
- Authentication token
- Channel whitelist/blacklist
- Response mode (DM/group chat)
Configuration Management
# 添加新渠道 openclaw channels add # 列出已配置渠道 openclaw channels list # 配对管理 openclaw pairing list openclaw pairing approve <platform> <code>Other extensions