LeanCTX

Tree-sitter can parse a 1,000-line Rust file into function signatures in under 1ms. That's the difference between 4,000 tokens and 200.

What's the first thing you set up when starting AI coding on a new project? I'm always curious how other devs handle the initial context.

npm downloads just crossed 18K/month on lean-ctx. Still a solo project. Still MIT licensed. Still can't believe this is real.

Crossed 860 GitHub stars on lean-ctx. Wild that something I built to fix my own token bill is now used by thousands of devs.

Unpopular opinion: context engineering matters more than prompt engineering. I've seen 70-95% token savings from smarter reads alone. Zero prompt changes.

The hardest part of building dev tools isn't the code. It's convincing developers they have a problem they didn't know they had.

When someone tells me they spend $200/month on AI tokens, my first question is always: Do you know what your agent is actually reading?

Today's debugging session: 2pm — "This should take 30 minutes" 5pm — "Why is tree-sitter returning empty ASTs for .vue files?" 8pm — Fixed. Shipped.

Honest question: how do you manage context when your AI session goes past 100K tokens? Building something for exactly this and I want to hear your workflow.

I used to think AI token costs would drop fast enough that optimization wouldn't matter. Then usage grew 300% while prices dropped 40%. Math doesn't lie.

Tree-sitter can parse a 1,000-line Rust file into function signatures in under 1ms. That's the difference between 4,000 tokens and 200.

Solo devs with good AI workflows ship faster than teams of 10 did in 2020. Context intelligence is the real multiplier.

When someone tells me they spend $200/month on AI tokens, my first question is always: Do you know what your agent is actually reading?

The wildest thing about building an open-source dev tool: Someone in Japan opened an issue at 3am my time. By morning, someone in Brazil had fixed it.

What's the first thing you set up when starting AI coding on a new project? I'm always curious how other devs handle the initial context.

Multi-agent workflows are the future. But each agent reads the same files independently. Without shared context, costs multiply linearly.

Confession: I sometimes spend more time curating what my AI reads than actually writing code. And honestly? My output has never been better.

Honest question: how do you manage context when your AI session goes past 100K tokens? Building something for exactly this and I want to hear your workflow.

I used to think AI token costs would drop fast enough that optimization wouldn't matter. Then usage grew 300% while prices dropped 40%. Math doesn't lie.

Writing code is getting easier every month. Understanding code is getting harder. The bottleneck shifted and most devs haven't noticed.