開発ツール
toon
TOON: a compact, human-readable serialization of JSON that saves tokens in LLM prompts
★ 25.5k スターMIT
ガイドを見る →headroomは、headroomlabs-aiが公開しているオープンソースの開発ツールプロジェクトです（GitHub: headroomlabs-ai/headroom）。GitHubスターは74,736、フォークは5,785で、Apache-2.0ライセンスの下で公開されています。GitHubでの最終更新は2026年10月8日です。
Compress tool outputs, logs and RAG chunks before they reach the LLM — same answers, fewer tokens
Decide what you want from headroom and check the README and requirements to confirm it fits your setup and Python environment.
Clone the repository from GitHub (github.com/headroomlabs-ai/headroom) or use the documented install channel if you do not need the source.
Follow the README quickstart to install dependencies and set any required configuration or credentials.
Work through the README's first example end to end, then adapt it to your own project or data.
Compresses tool outputs, logs, files and RAG chunks before they enter the context window.
The project reports 60-95% fewer tokens for JSON payloads while keeping answers the same.
Usable as a Python/TypeScript library, a proxy, or an MCP server.
Works with coding agents such as Claude Code and Cursor and frameworks like LangChain.
`headroom learn` adapts compression to your workload, per the README.
The proxy is built on FastAPI for easy self-hosting.
Start by getting the source and entering the project directory.
git clone https://github.com/headroomlabs-ai/headroom.git
cd headroomThe README recommends a uv tool install for the CLI, or pip for the Python package.
pip install "headroom-ai[all]"Start the proxy or MCP server and point your agent at it, per the 60-second get-started guide.
headroom # see README for the exact commandIt is maintained by headroomlabs-ai and contributors on GitHub, where the repository github.com/headroomlabs-ai/headroom is the canonical home for code, issues and releases.
Yes — it is open source under the Apache-2.0 license. Read the license in the repository for what is permitted, including any conditions on redistribution or commercial use.
No key of its own — it sits in front of your existing LLM provider, which still uses your own key.
Compared with manual context pruning: Headroom compresses content automatically as a proxy or MCP server instead of developers hand-trimming every payload.
Clone github.com/headroomlabs-ai/headroom, follow the README quickstart to install it, and run the documented first example before adapting it to your own work.
TOON: a compact, human-readable serialization of JSON that saves tokens in LLM prompts
Fast, accurate semantic code search for agents — about 99% fewer tokens than grep plus read.
Turn domain documents into high-quality LLM fine-tuning, RAG and evaluation datasets through a visual UI.
One Rust CLI for every LLM: shell assistant, chat REPL, RAG, tools and agents across 20+ providers.
⚡️next-generation personal AI assistant powered by LLM, RAG and agent loops, supporting computer-use, browser-use and coding agent, demo: https://demo.openag…
A free course to get into large language models, with roadmaps and runnable Colab notebooks
LibreAIカタログに掲載された開発ツールプロジェクトの中で最もスターの多いもの — headroomが合わなかった場合に検討したい、現実的な候補リストです。
The open source coding agent
The model-definition framework behind modern AI — thousands of pretrained transformers for text, vision, and audio.
Anthropic's agentic coding tool that lives in your terminal and works across your whole codebase
Cross-platform desktop all-in-one manager for Claude Code, Codex, OpenCode and other coding agents
OpenAI's lightweight coding agent that runs locally in your terminal
カタログの中でheadroomとスター数が近い開発ツールプロジェクト — 前後の棚もそのまま見て回れます。
Unified, efficient fine-tuning of 100+ LLMs and VLMs with a CLI and Web UI — ACL 2024
Spec-driven development (SDD) for AI coding assistants.
CLI proxy that cuts LLM token consumption 60-90% on common dev commands — one Rust binary
Autonomous coding agent available as an SDK, IDE extension and CLI assistant
A free course to get into large language models, with roadmaps and runnable Colab notebooks
OmO: a batteries-included orchestration layer that turns OpenCode into a multi-agent master