Cut 66%–96% LLM token costs for AI Agents with Thrift (Rust engine):
• 66–68% direct reduction on stack traces & git diffs • 95–96% cost savings on re-reads (SeenLedger) • 100% accuracy on Claude 3.5, GPT-4o, Gemini 2.5 Pro
Zero-alloc Rust core. DM for private CLI binary.
My (CLI) MCP and I against Anthropic’s wealth
2 days ago, I switched to Opus 5 because I think it will save more tokens for complex tasks than Sonnet 5. It burns tokens like crazy and still gets things wrong.
So I created personal trainers for Claude.