Topics / #context-window

Context-window projects on GitHub

Techniques for managing and compressing LLM context windows.

As of , the most-starred Context-window projects on GitHub are headroom (75.6K stars), context-mode (30.7K stars) and 12-factor-agents (26.8K stars). GitHub.Locker tracks 14 Context-window repositories with 162.4K stars combined. The fastest-growing on October 17 was context-mode, with +252 stars in a day.

14Projects
162.4KStars in total
+720Stars in a day
Browse in the catalog →github.com/topics/context-window ↗

Hot right now

Most stars gained in a day · Oct 17

+252

context-mode

mksglu/context-mode · ★ 30.7K

Context Mode is an MCP server plus hooks that keeps bulky tool output out of your AI coding agent's context window (98% reduction), persists session memory, and enforces tool routing across 17 platforms.

+213

billion-context

ranxianglei/billion-context · ★ 1.9K

billion-context is a context-compression proxy that sits between any coding agent and its model API, rewriting Anthropic/OpenAI streams so a 100K context window is enough and a single session can run for months across billions of tokens.

+78

headroom

headroomlabs-ai/headroom · ★ 75.6K

Headroom compresses everything an AI agent reads - tool outputs, logs, RAG chunks, files, and conversation history - before it reaches the LLM, cutting tokens while keeping the same answers. It ships as a library, a local proxy, and an MCP server.

+77

my-claude-code-mods

YeonwooSung/my-claude-code-mods · ★ 1K

A personal collection of custom 'mods' (plugins) for Claude Code that add things like performance profiling, context-waste analysis, a branching edit timeline with snapshots, repeated-failure detection, pinned memos, a model-effort indicator, and a small terminal game, built by YeonwooSung.

The all-time top

14 projects · by stars

  1. #1 · headroomlabs-ai/headroom

    headroom

    Headroom compresses everything an AI agent reads - tool outputs, logs, RAG chunks, files, and conversation history - before it reaches the LLM, cutting tokens while keeping the same answers. It ships as a library, a local proxy, and an MCP server.

    75.6K
  2. #2 · mksglu/context-mode

    context-mode

    Context Mode is an MCP server plus hooks that keeps bulky tool output out of your AI coding agent's context window (98% reduction), persists session memory, and enforces tool routing across 17 platforms.

    30.7K
  3. #3 · humanlayer/12-factor-agents

    12-factor-agents

    A written guide of 12 (plus a bonus 13th) engineering principles for building LLM-powered agent software reliable enough for production customers, distilled from the author's experience at HumanLayer and conversations with many SaaS founders.

    26.8K
  1. 4

    fast-jev-compaction

    7.7Ktamaratran/fast-jev-compaction · TypeScript

    A Claude Code plugin, also usable as an npm library, that replaces context compaction's summary with Jev decisions: every tool call and result is scored in fast requests, stale ones are dropped or truncated, and everything kept stays verbatim.

  2. 5

    pxpipe

    7.5Kteamchong/pxpipe · TypeScript

    pxpipe is a local proxy that cuts Claude Code's (and other LLM clients') input token usage by rendering bulky context — system prompts, tool docs, and conversation history — as compact PNG images instead of raw text, since an image's token cost depends on its pixel size, not the amount of text packed into it.

  3. 6

    claude-devtools

    4Kmatt1398/claude-devtools · TypeScript

    claude-devtools is a free, open-source debugging tool for Claude Code that reads the session logs already saved on your machine and reconstructs tool calls, thinking steps, subagent trees, token usage, and context window breakdowns in a visual desktop UI.

  4. 7

    MSA

    3.5KEverMind-AI/MSA · Python

    MSA (Memory Sparse Attention) is an end-to-end trainable, scalable latent-memory framework that lets LLMs handle context windows up to 100M tokens, replacing RAG or fixed-size RNN-style memory with a differentiable sparse-attention memory bank.

  5. 8

    dsh-context

    2.3Kbowenliang123/dsh-context · TypeScript

    dsh-context is a plugin for DeepSeek Harness (dsh) that gives full visibility into an AI agent's context window — a dashboard tab, a sidebar panel, and a /context slash command showing token stats, context composition, trends, compaction/prune events, file activity, and the agent family tree.

  6. 9

    billion-context

    1.9Kranxianglei/billion-context · TypeScript

    billion-context is a context-compression proxy that sits between any coding agent and its model API, rewriting Anthropic/OpenAI streams so a 100K context window is enough and a single session can run for months across billions of tokens.

  7. 10

    my-claude-code-mods

    1KYeonwooSung/my-claude-code-mods · TypeScript

    A personal collection of custom 'mods' (plugins) for Claude Code that add things like performance profiling, context-waste analysis, a branching edit timeline with snapshots, repeated-failure detection, pinned memos, a model-effort indicator, and a small terminal game, built by YeonwooSung.

  8. 11

    Spark-X2.5

    812XHToken/Spark-X2.5

    Spark-X2.5 is a series of two open-source language models (4B and 1.7B parameters) built for everyday conversation, writing, translation, coding, reasoning and agentic tool-use tasks, aiming for strong performance in a compact size.

  9. 12

    pi-posthorse

    272fitchmultz/pi-posthorse · TypeScript

    Posthorse is an extension for the Pi coding agent (specifically the fitchmultz/pi fork) that handles context window rollovers without summarization: when a session nears its limit it starts a fresh window with a durable handoff and a fully recoverable transcript instead of compressing the history into a summary.

  10. 13

    squeez

    217claudioemmanuel/squeez · Rust

    Squeez is a hook-based token compressor for seven AI coding CLI hosts (Claude Code, Copilot CLI, OpenCode, Gemini CLI, Codex CLI, Pi, Hermes) that intercepts tool output and shrinks bash results by up to 95% without losing exact identifiers.

  11. 14

    Iris

    102AllSpark-Research/Iris · Python

    Iris is a pair of open-weight search agent models (Iris-mini, 35B-A3B, and Iris-pro, 397B-A17B) post-trained from the Qwen3.5/3.6 series to decide what to search, how to read results, when to keep searching, and when enough evidence has been gathered.

Often tagged together