Configuration

CodeBuddy is configured through your editor’s settings. Open settings with Cmd+, (macOS) or Ctrl+, (Windows/Linux) and search for “CodeBuddy”.

AI provider setup

CodeBuddy supports 10 AI providers through a unified buildChatModel() factory. Configure your preferred provider:

Cloud providers

Provider	Setting key	Models
Anthropic	codebuddy.anthropic.apiKey	Claude Sonnet 4, Claude Opus 4, Claude Haiku
OpenAI	codebuddy.openai.apiKey	GPT-4o, GPT-4o-mini, GPT-4-turbo, o3-mini
Google	codebuddy.google.apiKey	Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.0 Flash
Groq	codebuddy.groq.apiKey	Llama 3.3 70B, Llama 3.1 8B, Mixtral 8x7B
DeepSeek	codebuddy.deepseek.apiKey	DeepSeek Chat, Coder, Reasoner
xAI	codebuddy.xai.apiKey	Grok models
Qwen	codebuddy.qwen.apiKey	Qwen Plus, Turbo, Max
GLM	codebuddy.glm.apiKey	GLM-4 Plus, GLM-4 Flash

Local providers

Provider	Setting key	Notes
Ollama	codebuddy.ollama.baseUrl	Default: http://localhost:11434. Any locally hosted model.
OpenAI-compatible	codebuddy.local.baseUrl	LM Studio, vLLM, or any server implementing the OpenAI API.

All providers share the same ChatModel type, so switching providers requires only a configuration change.

Key settings

{
  // Primary AI model
  "codebuddy.defaultModel": "claude-sonnet-4-20250514",

  // Auto-approve safe operations (file reads, searches)
  "codebuddy.autoApprove": true,

  // Maximum cost per task (USD, 6 decimal precision)
  "codebuddy.costLimit": 1.0,

  // Language for the UI (en, es, fr, de, ja, zh-cn, yo)
  "codebuddy.language": "en"
}

Agent limits

Control how far the agent can go before stopping:

{
  // Maximum events per task (default: 2000)
  "codebuddy.agent.maxEvents": 2000,

  // Maximum tool invocations per task (default: 400)
  "codebuddy.agent.maxToolInvocations": 400,

  // Maximum task duration in minutes (default: 10)
  "codebuddy.agent.maxDuration": 10,

  // Maximum concurrent agent streams (default: 3)
  "codebuddy.agent.concurrentStreams": 3,

  // Context window size in tokens (4k-128k)
  "codebuddy.context.windowSize": 128000
}

Security configuration

Permission profiles

Set the active permission profile to control what tools the agent can use:

{
  "codebuddy.security.permissionProfile": "standard"
}

Profile	Capabilities
restricted	Read-only. No terminal, no file writes, no browser.
standard	Read/write with safe terminal commands. Dangerous patterns blocked. Default.
trusted	Full access with auto-approval.

Access control

Control who can use the agent in team environments:

{
  "codebuddy.security.accessMode": "open"
}

Mode	Behavior
open	No restrictions (default)
allow	Only listed users can access
deny	Listed users are blocked, others allowed

Credential proxy

For teams that need centralized API key management:

{
  "codebuddy.credentialProxy.enabled": true,
  "codebuddy.credentialProxy.rateLimit": 60
}

The proxy runs on localhost, injects API keys from the OS keychain, and rate-limits requests per provider.

Project rules

Create a .codebuddy/rules.md file in your project root to set project-specific instructions that CodeBuddy follows in every conversation:

# Project Rules

- Use TypeScript strict mode
- Follow the existing code style
- Always write tests for new functions
- Use pnpm instead of npm

CodeBuddy also checks .codebuddy/rules/index.md, .codebuddyrules, and CODEBUDDY.md as fallback locations. Rules are limited to a 2,000-token budget (approximately 8,000 characters) and truncated with a warning if exceeded.

See Project Rules for details.

MCP servers

Enable and configure MCP server connections:

{
  "codebuddy.mcp.enabled": true,
  "codebuddy.mcp.timeout": 30000
}

Server definitions go in .codebuddy/mcp.json. See MCP Integration for full configuration.

Environment variables

For CI/CD or team setups, configure providers via environment variables:

export CODEBUDDY_OPENAI_API_KEY="sk-..."
export CODEBUDDY_ANTHROPIC_API_KEY="sk-ant-..."
export CODEBUDDY_GOOGLE_API_KEY="AIza..."

Next steps

• Settings Reference — Complete list of all 60+ settings
• Security — Detailed security model documentation
• Quickstart — Run your first task