Hey Product Hunt! 👋 I'm Yves, a developer from Switzerland. I built LeanCTX because I was burning through $200+/month on AI coding API costs — and couldn't figure out why. When I finally looked at the token logs, I was stunned: my AI tools were reading the same files over and over, completely uncompressed. A 300-line TypeScript file costs ~2,000 tokens. Read it 15 times in a session? That's 30,000 tokens — for ONE file. Multiply by every git command, every directory listing, every shell output... a single coding session burns ~90,000 tokens on context alone. So I built a fix. LeanCTX is a single Rust binary that sits transparently between your AI tool and the LLM. Three compression layers: → Shell Hook: 90+ CLI patterns (git, npm, cargo, docker, kubectl…) compressed before the output reaches the LLM. 60-95% savings. → Context Server: 42 MCP tools with smart caching, 10 read modes, delta updates, entropy filtering, and cross-session memory. Re-reads cost ~13 tokens instead of 2,000. → Intelligence Protocols: CEP, CCP, and TDD make the AI respond more efficiently too. 8-25% extra savings on top. What makes it different from alternatives: • Works with 24+ AI tools — not just one editor • One-command setup: `lean-ctx setup` auto-detects everything • tree-sitter AST parsing for 18 languages — not regex hacks • Zero telemetry. No analytics. No network requests. Not now, not ever. • Cross-session memory — your AI remembers context across chats Install in 60 seconds: curl -fsSL https://leanctx.com/install.sh | sh && lean-ctx setup It's fully open source (Apache 2.0), free forever, no account required. I believe developer tools should respect your privacy and your wallet. Check your actual savings with `lean-ctx gain` — most devs see 70-90% token reduction from day one. Would love your feedback. Happy to answer any questions!

LeanCTX

Save 99% on AI coding tokens. One binary. Zero telemetry.

Save 99% on AI coding tokens. One binary. Zero telemetry.