About the Project

Inspiration ✨

While building APIs, we kept running into the same old problem: testing and debugging felt like juggling too many tools. One moment, we were in Postman sending requests. The next, we were digging through stack traces in the terminal. Then back to updating specs, and then checking if the backend matched the contract. It was fragmented, repetitive, and, honestly, a little exhausting.

So we asked ourselves:
What if all of this could happen in one place—right inside the developer’s local workflow?
That simple question became the spark for Fixit-AI.

What We Learned 📚

• The importance of tight developer feedback loops: the faster the cycle from “run test” → “see result” → “understand fix,” the better the coding experience.
  
• How OpenAPI specifications can act as a single source of truth—not just for design, but for testing and even generating mocks.
  
• The surprising power of local LLMs: running offline models like gpt-oss 20B gave us natural language explanations without relying on external services.
  
• Building robust tools isn’t just about automation—it’s about reducing mental context switching for developers.
  

How We Built It 🛠️

We started by integrating directly into the VSCode terminal and building a CLI tool: fixit-ai.
The core building steps were:

1. Parsing OpenAPI specs to auto-generate pytest + httpx tests.
   
2. Running those tests seamlessly against a locally running server.
   
3. Capturing failures, then using a local LLM to explain errors in plain English and propose minimal code/spec/test fixes.
   
4. Extending into mock server generation, so frontend development doesn’t stall on missing backend endpoints.
   
5. Wrapping results with beautiful CLI + HTML reports powered by Rich for clarity.
   

The MVP flow became delightfully consistent:

fixit init --fastapi main:app --base http://localhost:8000
fixit gen && fixit test && fixit fix && fixit sec

Built With

• jinja
• makefile
• python
• shell