OpenWalrusConfiguration
Configure OpenWalrus — walrus.toml, directory structure, models, channels, and MCP servers.
OpenWalrus is configured via a TOML file at ~/.openwalrus/walrus.toml. The daemon creates a default config on first run.
Directory structure
~/.openwalrus/
├── walrus.toml # Main configuration
├── walrus.sock # Unix domain socket
├── agents/ # Agent definitions (*.md)
├── skills/ # Skill definitions
├── data/ # Memory database
└── work/ # Sandbox for file/bash operationsConfiguration file
Models
Set the default model and configure providers:
[model.default]
text = "deepseek-chat"
[model.providers.deepseek-chat]
model = "deepseek-chat"
api_key = "sk-..."Model names are auto-detected by prefix — gpt-* routes to OpenAI, claude-* to Anthropic, etc. See providers for the full list.
Channels
Connect messaging platforms:
[channel.telegram]
token = "your-bot-token"
[channel.discord]
token = "your-bot-token"
guild_id = "123456"
channel_id = "789012"See Telegram and Discord for setup details.
MCP servers
Register external tool servers:
[mcp_servers.my-server]
command = "npx"
args = ["-y", "@my/mcp-server"]See MCP integration for details.
Work directory
Override the sandbox directory:
work_dir = "/path/to/custom/work"Environment variables
API keys can use environment variable references in the config:
api_key = "${DEEPSEEK_API_KEY}"What's next
- CLI reference — all commands and flags
- Full config reference — every configuration field