yvgude/lean-ctx v2.9.0

v2.9.0 — Intelligence Layer + Pro Adaptive Intelligence

on GitHub

What's New

Intelligence Layer Rebranding

lean-ctx is now positioned as The Intelligence Layer for AI Coding — not just an MCP tool, but the compression layer between your AI and the LLM. Website, README, and all messaging updated.

Pro Adaptive Intelligence ($9/mo)

• lean-ctx upgrade — one-command Pro subscription via Stripe
• Adaptive compression models trained on community data
• Everything automatic: models sync daily, stats upload in background
• Cloud dashboard at leanctx.com/dashboard

Critical Fixes

• #29 — lean-ctx setup no longer overwrites existing editor configs (OpenCode, Cursor, Zed, VS Code) when JSON parsing fails
• #28 — lean-ctx read handles non-UTF-8 files gracefully (lossy decoding)
• Windows update — deferred binary replacement when MCP server locks the executable
• Compression regression — static rule-based prediction fallback for all users
• Cost constants — unified to $2.50/M input + $10/M output everywhere

New Features

• BM25 semantic code search (ctx_semantic_search, ctx_intent)
• Information Bottleneck filter for scientific compression
• MCP Unified Mode (LEAN_CTX_UNIFIED=1) — 24→5 tools, saves ~18K tokens
• mypy compression pattern + pytest direct routing
• Cloud backend with 12 API endpoints (auth, stats, Pro models, Stripe)

Install / Update

# Update existing installation
lean-ctx update

# Or fresh install
curl -fsSL https://leanctx.com/install.sh | sh
brew upgrade lean-ctx
cargo install lean-ctx
npm update -g lean-ctx-bin

Binary

• lean-ctx-darwin-arm64 — macOS Apple Silicon (this release)
• Windows and Linux binaries available via cargo install lean-ctx

Full changelog: https://github.com/yvgude/lean-ctx/blob/main/CHANGELOG.md

Full Changelog: v2.8.2...v2.9.0