LLM Explainer

The explainer module operates at the boundary of semantic source code and human language.

Role

Traditional reverse-engineering tools output massive dependency graphs that are impossible for humans to consume. aut0arch provides a higher-level abstraction. Once the clustering engine breaks the dependency graph into sub-systems, the explainer translates the code chunks back into human semantics.

It iterates through each generated cluster and collects the source code content of all encompassed files. It feeds this bulk text block directly to an AI language model alongside a finely-tuned system prompt asking for a high-level architectural overview.

By default, the explainer uses Google Gemini (gemini-flash-latest), but it also natively supports running locally via Ollama (e.g., gemma3:270m, qwen3.5:2b) for users who need enhanced privacy or offline capabilities.

Supported Models

• Gemini (Cloud): Fast, large context window. Requires `GOOGLE_API_KEY` in the `.env` file.
• Ollama (Local): Privacy-focused, runs entirely on your machine. Ensure an Ollama server is running locally (defaulting to `http://localhost:11434` via `OLLAMA_HOST`).

You can select the active backend during execution: ```bash

Run with Gemini (Default)

python explainer.py –model gemini

Run with local Ollama

python explainer.py –model ollama ```

LLM Outputs

The `explainer` forces the selected AI model to respond with a strictly typed JSON structure containing:

• A descriptive English title for the module
• An English paragraph describing the semantic role of the code files
• Deduced inputs and outputs to the module
• Specific code evidence identifying interesting patterns found within the provided context window

This information is integrated directly with the raw generated graph structures and output into a unified `clustering_data.json` for the frontend.