billion-context on GitHub: billion-token sessions in a 100K window, 5x token savings
aigclink · x · 2026-10-24
The billion-context GitHub repo (2.5k stars, 218 forks) ships a context-compression plugin targeting month-long single sessions of billions of tokens, small windows (100K is enough), and 5x token savings — "billion-context is all you need."
Core mechanism: incremental hierarchical compression plus on-demand recovery — agents fold old messages into summaries, restore originals via decompress, and search compressed history via searchcontext, while keeping the prompt cache prefix stable. The repo includes a paper, devlog, proof, and plugin adapters for Claude, Codex and more. MIT-licensed, installable via npm install -g billion-context.
Related event: billion-context: Run Billion-Token Sessions in a 100K Window(2 posts)→
More from coding & agent
- Dev runs T3 Code from his phone, spins up 3D modules in under a minute — gnukeith · 2026-10-24
- Agent harness was silently breaking its own KV cache, wasting 43% of a local LLM run — Stunning-Sherbet1853 · 2026-10-24
- Inference-time abliteration engine for llama.cpp ported to Strata, no finetuning needed — AccountAntique9327 · 2026-10-24
- Solo dev builds AI Dungeon Master RPG over 3 years, simulation engine comes before narration — tschilpi · 2026-10-24
- Cap pays GitHub sponsors for perf PRs: 452 merged, $1,275 to 18 devs — JosephJacks_ · 2026-10-24
- Trail of Bits open-sources a Claude Code skill to find hidden vulnerable Brotli copies (CVE-2026-111497) — banteg · 2026-10-24