Skip to content

AI Configuration

Configure AI models, providers, and custom agents.

Model Selection

Browse and select from available AI models:

  1. Go to Settings > Models
  2. Browse models by provider
  3. Filter by capability (chat, code, vision)
  4. Click a model to set as default

Model Capabilities

Capability Description
Chat General conversation
Code Code generation and analysis
Vision Image understanding
Function Calling Tool use support

Changing Models Mid-Session

You can switch models during a chat session:

  1. Click the model name in the chat header
  2. Select a different model
  3. Continue chatting with the new model

Context is preserved when switching models.

Provider Credentials

Configure API keys or OAuth for AI providers.

API Key Method

  1. Go to Settings > Provider Credentials
  2. Select a provider (OpenAI, Anthropic, etc.)
  3. Enter your API key
  4. Click Save

OAuth Method

For providers that support OAuth (Anthropic, GitHub Copilot):

  1. Go to Settings > Provider Credentials
  2. Select a provider with the OAuth badge
  3. Click Add OAuth
  4. Choose authorization method:
  5. Open Authorization Page - Opens browser for sign-in
  6. Use Authorization Code - Provides code for manual entry
  7. Complete the authorization flow

Testing Credentials

After adding credentials:

  1. Go to Settings > Models
  2. Models from that provider should appear
  3. Select a model and try sending a message

If models don't appear, verify your API key and check for errors.

Custom Agents

Create specialized AI agents with custom configurations.

Creating an Agent

  1. Go to Settings > Agents
  2. Click Create Agent
  3. Configure:
  4. Name - Display name for the agent
  5. Description - What this agent does
  6. System Prompt - Instructions for the AI
  7. Default Model - Model to use

  8. Allowed Tools - Which MCP tools it can access

  9. Click Save

System Prompt Tips

Write effective system prompts:

You are a code review expert specializing in TypeScript and React.

When reviewing code:
1. Check for type safety issues
2. Look for potential bugs
3. Suggest performance improvements
4. Ensure consistent code style

Be concise but thorough. Prioritize issues by severity.

Tool Permissions

Control which MCP tools an agent can use:

  • All Tools - Access to everything
  • Selected Tools - Only specific tools
  • No Tools - Pure conversation, no tool access

This is useful for:

  • Security-focused agents that shouldn't modify files
  • Research agents that only need read access
  • Specialized agents for specific tasks

Using Custom Agents

  1. Start a new session
  2. Click the agent selector
  3. Choose your custom agent
  4. Chat with the specialized agent

Context Usage

Monitor token usage with the context indicator:

  • Progress bar shows current usage
  • Updates as conversation grows
  • Warning when approaching limits

Managing Context

When context is running low:

  1. Use /compact to summarize history
  2. Start a new session with /new
  3. Be more concise in prompts
  4. Remove unnecessary file mentions

Context Limits by Model

Different models have different context limits:

Model Context Limit
GPT-4 8K - 128K tokens
Claude 3 200K tokens
GPT-3.5 4K - 16K tokens

Check your model's documentation for exact limits.