Skip to content

v0.8.0 — Context Intelligence System

Choose a tag to compare

View all tags
@JubaKitiashvili JubaKitiashvili released this 15 Mar 20:43
· 141 commits to main since this release
v0.8.0
6c321b8

Context Intelligence System — 99% Token Savings

ERNE v0.8.0 introduces a benchmark-verified context optimization system that compresses tool outputs by 97–100%, preserves exact code examples via FTS5 search, and leaves 99.6% of your context window free for problem solving.

Highlights

Layer What it does Savings
Content Summarizer Auto-detects 14 content types (docs, JSON, logs, test output, build output, CSV, git history) and compresses to statistical summaries 97–100%
Index + Search FTS5 BM25-ranked retrieval returns only relevant chunks — code examples preserved exactly 80%
Full Session Combined across a real debugging session (docs, snapshots, issues, tests, builds) 99%

Real numbers: 537 KB of tool outputs → 2.6 KB of context. See BENCHMARK.md for the full 21-scenario breakdown.

What's New

  • Content summarizer with 14 auto-detected content types
  • Content store with FTS5 BM25 search, Porter stemming, heading-based chunking
  • 21-scenario benchmark suite verified against real tool outputs (Context7, Playwright, GitHub, vitest, tsc, nginx, git)
  • Agent Preloader wiring — transition tracking with /api/context/predict endpoint
  • Hook execution metrics — duration, profile, and exit code reported to dashboard
  • PostToolUse context tracking — context_bytes in event flow

Install / Upgrade

npx erne-universal init

Full changelog: https://github.com/JubaKitiashvili/everything-react-native-expo/blob/main/CHANGELOG.md

Assets 2
Loading