Tilth – I spent tokens so my agents would stop wasting them (~4k Rust)
I'm an "ideas person" who messes around with AI on a low budget. I got tired of watching my tokens vanish and context windows filling up while agents fumbled around trying to find the right thing. Agents don't flail like they used to with shell tools, but there are still weak/blind spots and back-and-forth episodes — especially when using tools in combination/sequence. So I built "tilth" today. Or rather, AI built it — every line is Opus 4.6. I spent a lot of my precious tokens getting it to "not shit" (at least several of the different vendors' AI overlords assure me it's not…
In plain words
Tilth is a code-reading tool designed to help AI agents work more efficiently with limited token budgets. It provides code-aware summaries using tree-sitter to generate outlines of function signatures, types, and imports instead of raw file text, allowing agents to drill into specific sections as needed. The tool reduces wasted tokens and context window bloat during agent operations, particularly when combining multiple tools in sequence. Built entirely in Rust using Claude Opus, it targets developers and AI enthusiasts working with budget constraints.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
I'm an "ideas person" who messes around with AI on a low budget. I got tired of watching my tokens vanish and context windows filling up while agents fumbled around trying to find the right thing. Agents don't flail like they used to with shell tools, but there are still weak/blind spots and back-and-forth episodes — especially when using tools in combination/sequence. So I built "tilth" today. Or rather, AI built it — every line is Opus 4.6. I spent a lot of my precious tokens getting it to "not shit" (at least several of the different vendors' AI overlords assure me it's not shit). It's ~3,900 lines of Rust. I'm positive HN will let me know if its shit after all. The idea: give agents code-aware reads with a token budget. Read a large file and you get a tree-sitter outline (function signatures, types, imports) instead of the raw text. Drill into specific line ranges when you need detail. Search for a symbol and definitions come first (via AST), ranked by proximity to what you're editing. read — structural outline for large files, full content for small ones, line ranges to drill in search — symbol (tree-sitter AST), literal text, or regex. Definitions before usages. map — codebase overview in one call files — glob matching with token estimates, respects .gitignore session — tracks what the agent has looked at Tree-sitter adds structural awareness for 10 languages on top. Uses ripgrep internals for search, memmap for large files. Runs as a CLI or MCP server over stdio. Single binary, no indexing, no cloud. MIT licensed. To try it: cargo install tilth tilth map . tilth search "your_function_name" . tilth read src/main.rs tilth install claude-code I hope it helps other people who, like me, would rather spend their tokens building things than watching an agent navigate. -- *tilth* — the state of soil that's been prepared for planting. Good tilth means structured ground where things can take root. GitHub: https://github.com/jahala/tilth
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
