Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Global
#132 · +333 · 1d ·GitHub
Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.
#324 · +128 · 1d ·GitHub
上下文压缩插件，兼顾小窗口(100k上下文足矣)省token(省5倍token)和超长会话(数月级别几十亿token单会话)。billion-context is all you need
#621 · +61 · 1d ·GitHub
The best DeepSeek Harness plugin for context insight and management, with context dashboard / browser / sidebar and context command, for context statistics, composition, breakdown, evolution details, agent teams, understanding how the context is made of, and how it evolves.
#902 · +39 · 1d ·GitHub
npm test maps the core tests' claude-code/testing import to Node's test runner and Jest's expect; it does not emulate engine events. Files run sequentially so the existing performance assertions do not compete with other test processes. CI runs this independent core suite.
#1168 · +28 · 1d ·GitHub
What are the principles we can use to build LLM-powered software that is actually good enough to put in the hands of production customers?
#1520 · +20 · 1d ·GitHub
Claude Code plugin that replaces the compaction summary with Jev decisions: every tool call and result is scored in one fast request, stale ones are dropped or truncated, everything kept stays verbatim.
Cut Claude Code's input tokens by rendering bulky context as images — the same system prompt, tool docs, and history, in a fraction of the tokens.
#4576 · +3 · 1d ·GitHub
The missing DevTools for Claude Code — inspect session logs, tool calls, token usage, subagents, and context window in a visual UI. Free, open source.
Memory Sparse Attention - A scalable, end-to-end trainable latent-memory framework for 100M-token contexts.
#8029 · +0 · 1d ·GitHub
7B, two compact, general-purpose language models designed to make capable AI more practical, efficient, and accessible.
#8260 · +0 · 1d ·GitHub
An isolated Codex prototype tests local notes and recovery with Codex's native context reset. It is not ready for everyday use and is not included in the Pi package.
Hook-based token compressor for 7 AI CLI hosts (Claude Code, Copilot CLI, OpenCode, Gemini CLI, Codex CLI, Pi, Hermes). Up to 95% bash compression, signature-mode for code reads, cross-call dedup, MCP server, self-teaching protocol. Zero runtime deps.
6 series. A capable search agent has to decide what to search, how to read what comes back, when to keep going, and when the evidence it has gathered is enough. Iris is trained for exactly that loop.