Deebo

Deebo-Prototype: Your AI Agent's Debugging Copilot

What is Deebo-Prototype?

Deebo-Prototype is an agentic debugging copilot designed to assist AI coding agents in identifying and resolving complex bugs. It introduces a "multi-threaded" approach to the development workflow, significantly accelerating the time-to-resolution for debugging tasks.

How to use Deebo-Prototype?

Quick Install:

npx deebo-setup@latest

Manual Configuration: After installation, create a configuration file with server details for deebo-guide and deebo. Deebo is compatible with any OpenAI-compatible SDK, Anthropic, Gemini, and OpenRouter.

Debugging Session Workflow:

1. Start a Debugging Session: Instruct your AI agent to delegate debugging to Deebo, providing the error details (message, stack trace), the absolute path to your Git repository, and any helpful context (e.g., what you've tried, relevant files, reproduction steps, language/environment). Deebo will provide a session ID.
2. Check Investigation Progress: After about 30 seconds, ask your agent to check the session status using the session ID. You'll receive a "session pulse" with progress updates and findings.
3. Add Observations (Optional): Guide the investigation by providing observations to Deebo if you notice something important or think it's heading in the wrong direction.
4. Cancel a Session (Optional): If the issue is resolved or the investigation is no longer needed, instruct your agent to cancel the session.

Key Features of Deebo-Prototype

• Agentic Debugging: Acts as a copilot for AI coding agents, handling complex debugging tasks.
• Multi-Agent Hypothesis Testing: Utilizes isolated multi-agent hypothesis testing to fix hard bugs.
• Accelerated Debugging: Speeds up time-to-resolution by 10x.
• Compatibility: Works with OpenAI-compatible SDKs, Anthropic, Gemini, and OpenRouter.
• Session Management: Provides features to start, monitor, guide, and cancel debugging sessions.
• Memory Bank Access: Allows AI agents to access debug session logs for progress checking.

Use Cases of Deebo-Prototype

• Automated Bug Resolution: AI coding agents can delegate difficult bugs to Deebo for autonomous investigation and resolution.
• Enhanced Developer Productivity: Teams using AI agents for production code can maximize their productivity by leveraging Deebo's accelerated debugging capabilities.
• Complex Error Handling: Ideal for situations where AI agents encounter challenging or persistent errors that require in-depth analysis.

FAQ about Deebo-Prototype

• How does Deebo speed up debugging? By introducing a "multi-threaded" approach to the development workflow, allowing for isolated multi-agent hypothesis testing.
• What information should I provide to Deebo to start a debugging session? The error (message, stack trace, or behavior), the absolute path to your Git repository, and any helpful context.
• Can I guide Deebo during an investigation? Yes, you can provide observations to shift the direction of the investigation.
• What AI models does Deebo support? Deebo works with any OpenAI-compatible SDK, Anthropic, Gemini, and OpenRouter.
• Where can I find more information or support? Check the Deebo Guide or contact the developer on X (@sriramenn).

Deebo: Your AI Agent's Debugging Copilot

Deebo is an agentic debugging copilot for your AI coding agent that speeds up time-to-resolution by 10x. If your main coding agent is like a single-threaded process, Deebo introduces multi-threadedness to your development workflow.

feedback, questions/support? check out Deebo Guide below, or dm me on x @sriramenn

If you think your team can benefit from Deebo, we’d love to hear from you. We’re partnering with teams who use AI agents to write production code and want to maximize their productivity. Reach out for a live walkthrough, custom setup support, or to explore early access to enterprise features.

Quick Install

npx deebo-setup@latest

After installing with deebo-setup, create a configuration file at your coding agent's specified location with the following content. First, add the guide server (which provides help documentation even if the main installation fails):

{
  "servers": {
    "deebo-guide": {
      "command": "node",
      "args": [
        "--experimental-specifier-resolution=node",
        "--experimental-modules",
        "/Users/[your-name]/.deebo/guide-server.js"
      ],
      "env": {},
      "transportType": "stdio"
    },
    "deebo": {
      "command": "node",
      "args": [
        "--experimental-specifier-resolution=node",
        "--experimental-modules",
        "--max-old-space-size=4096",
        "/Users/[your-name]/.deebo/build/index.js"
      ],
      "env": {
        "NODE_ENV": "development",
        "USE_MEMORY_BANK": "true",
        "MOTHER_HOST": "openrouter",
        "MOTHER_MODEL": "anthropic/claude-3.5-sonnet",
        "SCENARIO_HOST": "openrouter",
        "SCENARIO_MODEL": "deepseek/deepseek-chat",
        "OPENROUTER_API_KEY": "your-openrouter-api-key"
      }
    }
  }
}

Deebo works with any OpenAI-compatible SDK, Anthropic, Gemini, and OpenRouter.

Deebo helps your AI agent debug real software errors by launching automated investigations. Here's how to use it effectively.

---

1. Start a Debugging Session

When you hit a tough bug, ask your agent to delegate the task to Deebo.

What to include in your request:

• 🔧 The error (message, stack trace, or behavior)
• 📁 The absolute path to your Git repository
• 💡 Any helpful context, such as:
  • What you’ve already tried
  • Relevant files or code snippets
  • How to reproduce the issue
  • The language or environment

Example instruction to your agent:

“This error is happening in /path/to/repo, possibly related to auth logic. I already checked the session token parser. Can you delegate this to Deebo?”

Your agent will start a Deebo session and give you a session ID (e.g. session-1745...). Save it.

---

2. Check Investigation Progress

After ~30 seconds, ask your agent to check the status of the Deebo session using that session ID.

You’ll get a session pulse, which shows:

• Whether the investigation is in progress or completed
• What the system is currently exploring
• Summaries of findings, if any

---

3. Add Observations (Optional)

If you notice something important — or think Deebo is heading the wrong way — you can guide the investigation.

Ask your agent to pass a short observation to Deebo.

Example:

“Let Deebo know that the file size warnings might be a red herring — the issue is probably with the CI env vars.”

This may shift the direction of the investigation.

4. Cancel a Session (Optional)

If you fixed the issue or no longer need the investigation, tell your agent to cancel the Deebo session.

For AI Agents: Memory Bank Access

When asked to check debug session progress, look in: ~/.deebo/memory-bank/[project-hash]/sessions/[session-id]/logs/

The project hash is a unique identifier for each repository, and session IDs are provided when starting a debug session.

Want More?

We're piloting enterprise features that unlock unprecedented productivity gains for your team. Reach out if interested!

---

Watch the full work session with Cline + Deebo here (3 mins, sped up)

License

Apache License 2.0 — see LICENSE for details.