context-compression

14-stage Fusion Pipeline for LLM token compression — reversible compression, AST-aware code analysis, intelligent content routing. Zero LLM inference cost. MIT licensed.

Not only memory but a full story.

Context compression for AI coding agents: compresses tool output before it enters the model, dedups repeats to 13-token refs. Claude Code, Cursor, Codex, Kiro, Zed, any MCP client. Rust, zero LLM calls.

Cut AI context cost without trusting the compressor. Every reduction is reversible, byte-exact recoverable, and carries an auditable receipt. Local-first, works through proxy, MCP, SDK, or agent wrapper.

Your agent pays twice for output it has already seen. OMNI returns a handle instead: 97.2% off a file read twice. Nothing deleted, nothing invented.

An effective and efficient coding-agent setup in under 30 seconds: best tools, unified instructions, any OS.

Official Implementation for the ACL 2026 paper "SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression"

Portable CC-inspired skills for memory, verification, multi-agent coordination, context compression, and proactive coding-agent workflows.

One zero-dependency CLI for every MCP server and agent skill. Token optimization, tool discovery and context compression: 71,929 -> 581 tokens (-99.2%, measured), schemas stay out of context. One config for Claude Code, Codex, Cursor, every agent. 341KB, pure Python. | 零依赖 CLI：管所有 MCP 工具与技能，工具发现省 99.2% token。

面向 AI Agent 的高并发、低延迟上下文治理与 Prompt Cache 守护网关 (Tree-sitter AST / Tool Delta / 50%~80% Token 削减)

Context compression for MCP. Same upstream call, exact results recoverable.

A drop-in proxy that compresses bloated code context in real-time, cutting LLM API costs by 50–80% without losing what the model actually needs to know.

The official repo for "LLoCo: Learning Long Contexts Offline"

Zed extension for Headroom — context compression for AI agents

Context compression plugin for Claude Code. Automatically trims large tool output—JSON, YAML, stack traces, and logs—before it enters the context window.

Pytorch implementation for "Compressed Context Memory For Online Language Model Interaction" (ICLR'24)

State aware knowledge compression, ingestion, and hybrid retrieval engine. Zero dependencies. Sub-100ms queries.

铃音的记忆写入协议 - an informatics protocol: fragmented, cue-based, symbolic, highly compressed information for AI agents; usable for memory writing and transmission, decodable approximately or via retrieval.

⚡ Cut Claude token usage by 90%+ — free, open-source, local-first context compression for Claude Code. Hybrid RAG (BM25 + ONNX vectors), AST chunking, reranking. No API needed.

Cursor uses AI to edit code — we use AI to edit AI's context. 🪆 Context map + compression + version control for LLM context windows.