Chat - General-Purpose Collaborative Thinking Partner

Your go-to tool for quick questions, brainstorming, and open-ended discussions with AI models.

Example Prompts 

"Ask the AI what the best approach is for implementing retry logic in this service"

"Chat with a model about the tradeoffs between SQL and NoSQL for our use case"

"Continue our previous conversation about the migration strategy"

"Look at these files and explain how the authentication flow works"

"Analyze this screenshot and describe what the UI layout is doing"

How It Works

Chat is a single-pass tool (uses createSimpleTool) that sends your prompt directly to an AI model through bab’s provider registry. Unlike workflow tools, it does not orchestrate multi-step investigations or expert validation – it gives you a direct line to the model for collaborative thinking.

  1. Validates the working directory exists on disk.
  2. Loads any specified files into context (with size limits enforced).
  3. Resolves the model through bab’s provider registry (Google, OpenAI, Anthropic, OpenRouter, or custom providers).
  4. Sends your prompt with the CHAT_SYSTEM_PROMPT and returns the model’s response.
  5. Tracks conversation history if you use continuation_id, allowing multi-turn exchanges.

Key Features

  • Multi-provider support – routes through bab’s provider registry to Google, OpenAI, Anthropic, OpenRouter, or custom endpoints.
  • Conversation continuation – maintain context across calls with continuation_id (max 8 turns of history, 16 total messages before a new continuation is required).
  • File embedding – attach files for the model to reference, with size limits (50k tokens per file, max 40% of the context window).
  • Image support – pass image file paths or base64 blobs for visual analysis tasks.
  • Working directory validation – the provided path must be an absolute path to an existing directory.
  • Temperature control – tune response creativity from deterministic (0) to highly creative (1).
  • Thinking mode – adjust the model’s reasoning depth from minimal to max.

Tool Parameters

  • prompt (string, required) – Your question or discussion topic.
  • working_directory_absolute_path (string, required) – Absolute path to an existing directory. Used as working context for the conversation.
  • continuation_id (string, optional) – ID from a previous chat response to continue that conversation thread.
  • absolute_file_paths (string[], optional) – List of absolute file paths to embed as context for the model.
  • images (string[], optional) – Image file paths or base64-encoded blobs for visual analysis.
  • model (string, optional) – Specific model ID to use (e.g., gemini-2.5-pro, gpt-4o). Defaults to the provider’s configured model.
  • temperature (number, optional) – Sampling temperature from 0 to 1. Lower values are more focused, higher values more creative.
  • thinking_mode (enum, optional) – Controls reasoning depth. One of: minimal, low, medium, high, max.

Usage Examples

Quick question 

{
  "prompt": "What are the pros and cons of using WebSockets vs SSE for real-time updates?",
  "working_directory_absolute_path": "/home/user/project"
}

File-aware discussion 

{
  "prompt": "Explain what this middleware does and suggest improvements",
  "working_directory_absolute_path": "/home/user/project",
  "absolute_file_paths": [
    "/home/user/project/src/middleware/auth.ts",
    "/home/user/project/src/middleware/rateLimit.ts"
  ]
}

Continuing a conversation 

{
  "prompt": "Now how would we handle the edge case where the token is expired?",
  "working_directory_absolute_path": "/home/user/project",
  "continuation_id": "abc123-previous-chat-id"
}

Visual analysis 

{
  "prompt": "What layout issues do you see in this screenshot?",
  "working_directory_absolute_path": "/home/user/project",
  "images": ["/home/user/project/screenshots/dashboard.png"]
}

Using a specific model with tuned parameters 

{
  "prompt": "Generate 5 creative names for a CLI tool that manages dev environments",
  "working_directory_absolute_path": "/home/user/project",
  "model": "gemini-2.5-pro",
  "temperature": 0.9,
  "thinking_mode": "low"
}

Best Practices

  • Start simple – most questions don’t need file attachments or special parameters. Just send a prompt.
  • Use file embedding sparingly – only attach files the model actually needs. Each file consumes context window budget.
  • Leverage continuation – for multi-turn discussions, always pass the continuation_id from the previous response rather than re-explaining context.
  • Watch the turn limit – conversations max out at 16 total messages (8 turns). Start a fresh continuation for long discussions.
  • Match the model to the task – use faster/cheaper models for simple questions, stronger models for nuanced analysis.
  • Set temperature intentionally – use low temperature (0-0.3) for factual questions, higher (0.7-1.0) for creative brainstorming.

When to Use vs Other Tools

Scenario Use Why
Quick question or brainstorm chat Single-pass, fast, direct model access
Deep multi-step investigation thinkdeep Structured workflow with expert validation
Code structure analysis analyze Purpose-built for codebase understanding
Debugging a specific error debug Specialized error diagnosis workflow
Reviewing code changes codereview Focused on diff review and quality
Need multiple models to agree consensus Cross-model validation

Chat is your everyday conversational tool. Reach for it when you need a quick, direct exchange with an AI model – no orchestration, no multi-step workflows, just a straightforward conversation.