DigitalOcean GitHub Copilot Extension

Running the Application

To start the application, follow these steps:

1. Ensure you have Python 3.10+ installed.
2. Install dependencies:

   pip install -r requirements.txt

3. Run the FastAPI application using Uvicorn:

   uvicorn server:app --host 0.0.0.0 --port 8000 --timeout-keep-alive 600 --log-level info

---

Setting Up the Forwarding URL

To integrate this extension with GitHub Copilot, follow these steps:

1. Go to Developer Settings in your GitHub App.

2. Paste the forwarding URL into the Agent URL under Copilot settings.

3. Obtain the URL from one of the following sources:

   • Ngrok: If using Ngrok, run:

     ngrok http 8000

     Copy the generated HTTPS forwarding URL and paste it into the Agent URL field.

   • App Platform: If deploying on DigitalOcean App Platform, use the assigned public URL.

Once configured, GitHub Copilot will send requests to your extension, enabling real-time responses within the Copilot chat UI.

Overview

We are developing a DigitalOcean Extension for GitHub Copilot, enabling users to interact directly with a DigitalOcean agent that specializes in:

• DigitalOcean documentation
• Product insights
• DO-specific recommendations

This extension is designed for users leveraging DigitalOcean resources and aims to provide relevant, DO-oriented guidance seamlessly within Copilot Chat.

Development Background

This extension is built upon concepts outlined in this tutorial, expanding it into a FastAPI-based application that integrates with DigitalOcean's agent system.

Initial Implementation

The original idea was to:

1. Connect with the DigitalOcean extension within GitHub Copilot Chat.
2. Forward queries (including attached code) to the DigitalOcean agent for processing.
3. Send the agent's response to GitHub’s API to regurgitate the exact DigitalOcean agent response.

Problem with GitHub Copilot's API

Copilot Contradicting External Agent Knowledge

• Initially, responses from the DigitalOcean agent were sent to GitHub Copilot's API.
• However, Copilot often contradicted the agent's responses, favoring its own training data over external sources.
• This defeated the purpose of integrating DigitalOcean’s expertise into Copilot Chat.

Alternative Approach

Replicating GitHub Copilot’s Streaming Response Format

• Instead of invoking GitHub's API, we emulated the response structure Copilot expects.
• By replicating Copilot’s streaming format, we successfully rendered responses in the Copilot Chat UI without relying on GitHub’s API.

Current Challenges

While short responses work for now, longer or complex queries (including attached full code) trigger errors, such as:

• “Internal agent error”
• “GitHub was unable to connect to ‘doextension’”

These indicate possible issues with:

1. Timeouts – GitHub Copilot may drop connections if responses take too long.
2. Rate Limits – Requests to the DigitalOcean agent might be exceeding thresholds.
3. Streaming Issues – The response stream might not be keeping Copilot’s connection alive properly.

Next Steps

• Understand what was triggering those error messages as we need to accommodate longer more complex queries for this extension to work with zero exceptions.
• Ensure immediate response streaming to keep the connection open while waiting for the DigitalOcean agent.
• Investigate timeouts and adjust FastAPI/Uvicorn settings to prevent premature disconnections.
• Optimize response chunking to handle longer DigitalOcean agent outputs without truncation.

Our goal is to seamlessly integrate the DigitalOcean agent into GitHub Copilot without API contradictions, ensuring fully rendered responses for complex queries.