Pithos — Cut your AI coding bill, locally

Pithos optimizes your prompts on your machine before they reach your AI coding tool (Claude Code, Cline, Aider, raw API). Smart context selection + prompt caching layout + output minimization + semantic cache → typically 40–70% fewer tokens, same quality.

Your code never leaves your machine.

Requirements

• The local Pithos engine running (bundled pithos-engine.exe, or python server/main.py, port 8765)
• No Ollama or external model server needed — embeddings run locally via fastembed; the multilingual model (~220MB) downloads automatically on first run.

Commands

• Pithos: Optimize et — scan the workspace, produce a lean prompt + show savings
• Pithos: Ayarlar — output mode, context budget, model, license
• Pithos: Kurulum sihirbazı — first-run setup

Privacy

All processing is local. Pithos never sends your code to any cloud service. In proxy mode it forwards requests to your AI provider using your own API key, which is never stored.

Learn more: https://pithosai.com