Prompt-History-Builder

A Python tool to distill raw chat logs into abridged prompt_history.txt files—unpacking how LLM-driven projects come to life at the edge of tech and purpose.

What It Does

• Turns chaotic chats into structured summaries (e.g., [User] 'prompt...' - summary) using local parsing or Venice.ai’s API.
• Tracks LLM-driven tool development—born from a pro-life card game built with Grok 3.
• Supports chunking for big logs—processes files exceeding model context sizes.

Why It’s Useful

• Debugging: Reveals how LLM projects (like this one) unfold and stumble.
• Learning: Teaches others (and future LLMs) to build with AI, step by step.
• Roots: Tied to stronghold-quest—pro-life tech at its peak (LLMs, Web3).

Setup

1. Clone: git clone https://github.com/EricTylerZ/prompt-history-builder.git
2. Install deps: pip install openai python-dotenv
3. Copy configs:
   • copy .env.example .env—add your Venice API key (from https://venice.ai/settings/api).
   • copy prompts.example.json prompts.json—edit prompts as needed.
   • copy models.example.json models.json—run python src/list_models.py to refresh.
4. Run:
   • Local: python src/main.py projects/your_chat.txt
   • Test with Venice: python src/test_main.py projects/your_chat.txt --models llama-3.2-3b --prompts p1

Structure

• src/:
  • main.py: Core tool with Venice API integration.
  • test_main.py: Testing arena—runs multiple models/prompts, handles chunking.
  • list_models.py: Fetches Venice model info to models.json.
  • manage_dirs.py: Sets up project dirs (output/, logs/).
  • prompt_history_builder.py: Original local parsing logic.
• projects/: Your workspace—empty in repo via .gitignore.
• output/test/: Test outputs—ignored.

Chunking

• For logs > model context (e.g., 131k tokens for llama-3.2-3b):
  • Splits into chunks (80% of context size).
  • Adds “Note: The below was chunked...” only when chunked—avoids clutter on small files.
• Run: python src/test_main.py projects/big_chat.txt --models llama-3.2-3b --prompts p3.

The Story

Started with a pro-life card game, built with Grok 3, and grew into a meta-tool for LLM projects. It’s about harnessing AI to track prompt history—think Git-managed logs for pro-life tech, with Web3 on the horizon—all to defend life through innovation.

Next Steps

• Test chunking on big logs: python src/test_main.py projects/full_chat.txt --models llama-3.3-70b --prompts p1.
• Contribute: Fork it, refine prompts in prompts.json, or tweak models—build with it!