Token Optimization: open-source projects on GitHub
33 repositories on Token Optimization, picked by the REDDYX AI agent. Each has a summary, key features and a breakdown. Sorted by stars.
// RELATED TOPICS
LLM Input Compression
Compresses tool outputs, logs, files, and RAG chunks before sending to the LLM, cutting token usage by 60‑95% without loss of quality.
★ 79,744 · PythonRTK: CLI proxy
RTK — CLI proxy reducing LLM token consumption by 60‑90 % on typical dev commands. One Rust binary with no dependencies.
★ 46,230 · RustOmniRoute: AI Gateway
Free AI gateway with a single endpoint and support for over 160 providers. Delivers up to 95% context compression, auto-fallback, and multimodal APIs for desktop and PWA.
★ 4,367 · TypeScriptheadroom — compression for LLM
Compresses tool outputs, logs, files, and RAG chunks before sending to the LLM, cutting tokens by 60–95% while preserving answer quality. Library, proxy, and MCP server.
★ 1,828 · PythonLean-ctx: OS Context for AI
Reduce token usage in Cursor, Claude Code, Copilot and other AI tools by 60–95% (up to 99% with caching). One Rust program with 49 tools, 10 read modes, and 90+ patterns.
★ 1,667 · RustMiddleware OpenWolf
More precise context and lower token consumption as open middleware for Claude Code.
★ 1,606 · TypeScriptPARITOK-4B: REDUCING TOKEN USAGE 90%
Lossless code compression gateway cuts AI agents' token usage by 25% per turn and up to 90%+ in long sessions. Open 4B model, ready to use with Claude Code, Cursor, Codex
★ 1,446 · PythonREDUCE AI TOKEN COSTS
Chisle cuts AI-agent token usage: terse prose, YAGNI-first code, and tool-output compression. Works with Claude Code, Pi, Cursor, Codex, Gemini and four more, zero deps,
★ 1,072 · JavaScriptMEMORY CLAUDE CODE IN OBSIDIAN
The project reduces tokens in Claude Code session by 71,5× via Obsidian + Graphify. It provides persistent memory, knowledge graphs of the codebase and chat import. Inter
★ 1,067 · PythonToken Optimizer
Identifies and removes redundant tokens. Improves context quality during compression and prevents data degradation.
★ 1,034 · PythonCONTEXT MANAGEMENT FOR CLAUDE!
The plugin for Claude Desktop and Codex allows you to select still which past images to send with the next message. Unmarked images still become placeholders, the conver
★ 803 · TypeScriptSQZ: COMPRESSION FOR AI AGENTS
SQZ compresses tool output before it reaches the LLM, replacing repeated fragments with short 13‑token references. Works with Claude Code, Cursor, Codex, Kiro, Zed and an
★ 677 · RustKNOWLEDGE GRAPH FOR AI AGENTS
Project turns codebase into knowledge graph grounded in formal methods. AI agents get exact code answers, ensuring correctness and saving tokens.
★ 675 · TypeScriptACCELERATING CODE REFACTORING
Tura reduces the number of code rewrite steps by 83.1% and increases DeepSWE pass rate by 16.7 percentage points compared to Codex CLI.
★ 667 · RustDUMPED TOKENS IN THE TRASH 🚀
Ready setup for Claude Code that cuts documentation costs by 90% in 5 minutes. Any project, one script — ⭐ 471 stars, JavaScript.
★ 656 · JavaScriptDOUBLE CLOUD CODE
headroom-desktop written in Rust lets you double the usage limits of Claude Code and Codex. More agents, fewer restrictions.
★ 646 · RustTOKEN OPTIMIZER: -95% TOKENS
Reduces token usage in Claude Code by 95%+ through caching, compression, and smart tool selection.
★ 582 · TypeScriptRATEL SMART MEMORY FOR AGENTS
Ratel is a Rust framework for AI‑agent context engineering: it cuts token usage by ~80% and provides in‑process BM25 plus semantic search without a vector DB.
★ 563 · RustLOWFAT - TRIMS COMMAND NOISE
Rust utility that removes noise from command output and saves tokens. Ideal for AI agents and developers.
★ 478 · RustMCPTOON: SAVES 99% OF MCP TOKENS
Standard MCP discovery burns 71,929 tokens. mcptoon drops it to 581 — 99.2% savings. Zero dependencies.
★ 431 · HTMLENTROLY - 80% SAVINGS ON AI
Entroly compresses AI context, detects hallucinations, and saves up to 80% on API costs for OpenAI, Anthropic, and Gemini.
★ 402 · PythonTokeless Fast Agent Coding
Set up a powerful AI coding agent in 30 seconds: best tools, unified guide, support for all OS.
★ 380 · GoCONTEXERE: CONTEXT LAYER FOR AI
Contexere (formerly OMNI) — context layer for AI coding agents. Compresses context by removing noise and saving tokens without information loss. Written in Rust, works vi
★ 377 · RustHONEY: Fewer Tokens, More Results
Honey reduces AI agent token usage by 53% without quality loss, working with Claude Code, Cursor, Copilot and others.
★ 356 · JavaScriptLAP: Agent API Compass
LAP converts any API specifications into a compact agent‑native format, reducing size by 10× and providing ready‑to‑use skills for 1500+ APIs.
★ 349 · PythonSNIP - TOKEN SAVINGS UP TO 90%
Go-based CLI proxy that reduces LLM token usage by 60-90% with YAML filters for Claude Code, Cursor, Copilot, and Gemini.
★ 280 · GoLLMTRIM: LLM Request Compression
A local proxy that cuts tokens in LLM requests without altering responses, saving up to 31% of input and 74% of output tokens.
★ 274 · RustTOMC: CONTEXT AND MEMORY FOR LLM
TOMC compresses CPU context and stores task memory for Codex, Claude Code, Cursor, and other MCP clients. Extends context window without quality loss. 204★ on GitHub.
★ 264 · PythonSAVES 50K TOKENS VIA INDEX
Compact codebase index for AI assistants — saves 50K+ tokens per conversation. TypeScript.
★ 262 · TypeScriptSAVE TOKENS IN CLI
Local CLI optimaizr finds wasted tokens in Claude Code, Codex, opencode and your API calls. Analyzes spending and shows where money is lost. Supports TypeScript projects
★ 243 · TypeScriptAGENTIC OS: MANAGING AI-AGENTS
Agentic OS - a framework for managing AI agents that runs them through a five-step workflow: plan, build, review, test and release. Each step is considered complete only
★ 236 · PythonLLM TOKENOMICS: TOOLS TO SAVE
Curated list on AI token economics: costs, waste, and how to cut bills. Tools, benchmarks, papers, and ready configs for LLMs and coding agents. 206 stars on GitHub.
★ 219 · PythonSEMBLE: Code Search for AI Agents
Hybrid BM25 + semantic search with AST and dependency analysis. A replacement for grep/cat/read/ls for Claude Code, Cursor, Aider, and other AI agents.
★ 63 · RustNew projects on this topic in Telegram
Telegram →