Configuration

The recommended way to configure CrabTalk is with crabtalk auth, which walks you through model setup interactively. On first run, crabtalk daemon start prompts you automatically.

You can also edit the configuration file directly at ~/.crabtalk/crab.toml.

Quick setup

crabtalk auth

This adds or edits model providers, API keys, and agent defaults. Run it again any time you want to change your configuration.

Configuration file

The file at ~/.crabtalk/crab.toml can be edited directly for full control. See the full example config for all available options.

Default agent

The [system.crab] section configures the daemon's default agent:

[system.crab]
model = "deepseek-chat"
thinking = false

The model field references a model name from a [provider.*] section.

Providers

Each provider gets a [provider.<name>] section with a models list:

[provider.deepseek]
models = ["deepseek-chat", "deepseek-thinking"]
api_key = "sk-..."

[provider.anthropic]
models = ["claude-sonnet-4-20250514"]
api_key = "sk-ant-..."
standard = "anthropic"

Supported standards: openai_compat (default), anthropic, google, ollama, azure, bedrock. See providers for the full list.

MCP servers

[mcps.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/docs"]

Agents

Define sub-agents in TOML (in addition to Markdown files in agents/):

[agents.researcher]
description = "Research specialist"
model = "claude"
max_iterations = 32
thinking = true
members = ["writer"]

What's next

Commands — gateway and search setup
Agents — agent configuration and delegation
MCP Servers — connect external tool servers