Release notes, architecture write-ups, and the benchmark work behind the numbers on the front page.
Terminal noise is what you are paying for. What semantic distillation cuts, what it does not, and the measured numbers behind both.
Nine claims OMNI could not support, fixed, and an all-time savings figure corrected downward because 73.4% of it was terminal output nobody is billed for.
A truncation that hides its own size is a lie by omission. v0.6.7 makes six of them state the count, and removes three more invented verdicts.
A self-audit found OMNI fabricating success and deleting the answer on find, ls and git log. v0.6.6 fixes seven of these and commits the benchmark.
For two releases OMNI reported a saving Claude Code had rejected: 27 of 34 Bash calls in one session. v0.6.5 fixes the payload shape, not just the key.
v0.6.4 adds a re-run metric reduction cannot fake, stops omni stats counting savings the host threw away, and fixes eleven ways OMNI could invent a result.
v0.6.3 stops OMNI distilling commands that already failed, stops distillers reading their own collapse markers as data, and fixes a corrupted exec.
v0.6.2 stopped OMNI corrupting JSON and YAML, made find lossless, and replaced every published number with a measured replay. 97.3% fell to 58.9%.
v0.6.1 rewrote the README around terminal noise and context amnesia. It also shipped a find distiller claiming 99.8% savings, which v0.6.2 retracted.
OMNI v0.6.0 introduces Autonomous Loop Engineering, Maker-Checker Verification, and critical UTF-8 panic resolution for production-grade Agentic AI.
OMNI v0.5.9 introduces Engrams for subtask digests, a Session Health Dashboard, Smart PreCompact v2, and Session Handoff capabilities.
OMNI v0.5.8 introduces the memory-efficient Streaming Distillation Pipeline, expanded semantic distillers for 9 developer tools, and advanced context analytics.
Why cutting output isn't enough. Discover how OMNI balances strict token efficiency with deep context fidelity for long-running AI sessions.
Raw terminal output beside OMNI's distilled signal across Cargo, Jest, Kubectl and Docker, with the cases where it saves nothing included.
A technical deep dive into the Rust-powered engine that intercepts, scores, and collapses your terminal output in sub-milliseconds.
OMNI v0.5.7 brings multi-agent awareness, persistent project knowledge, and an automated outer loop for LLM filter optimization.
OMNI v0.5.5 adds deep heuristic path analysis, cloud infrastructure classification, and update telemetry for context-aware token distillation.
OMNI v0.5.4 introduces session-aware distillation, three new core distillers, a modular pipeline architecture, and zero-allocation ANSI processing.
OMNI v0.5.2 adds native filter support for npm, yarn, pnpm, bun test runners and Python pytest variants, with improved test diff preservation.
OMNI v0.5.3 adds the omni update command, automated version checks from GitHub, safety prompts before destructive operations, and full hook diagnostics.
OMNI v0.5.1 ships omni reset for safe configuration removal, automated release workflows, omni learn stability fixes, and sub-50ms latency assertions.
OMNI v0.5.0 is a ground-up rewrite in Rust: a sub-5MB binary, zero startup overhead, SQLite WAL persistence, and session-aware distillation.
OMNI v0.4.5 delivers 60+ polyglot semantic filters, SHA256 hook integrity verification, project trust boundaries, and autonomous filter discovery.
OMNI v0.4.3 achieves perfect version alignment across all nine manifest and source files, eliminating package drift in multi-ecosystem releases.
OMNI v0.4.4 adds a full test suite, fixes shell injection holes, eliminates Wasm memory leaks, and introduces dynamic signal-density scoring.
OMNI v0.4.2 introduces a unified CLI design system, agent autopilot aliases for transparent interception, and fixes for the DSL engine and filters.
OMNI v0.4.1 adds MCP integration for Google Antigravity, proxy command distillation, real-world examples, and filesystem exploration tools.
OMNI v0.3.7 introduces a local metrics system, multi-period analytics reports, per-agent filtering, and critical installer and symlink fixes.
OMNI v0.3.8 achieves full version synchronization across eight files and upgrades the release automation pipeline for cross-platform reliability.
OMNI v0.3.9 exposes the semantic stripping engine to users via a JSON-based DSL, ships omni uninstall, and introduces dynamic confidence scoring.
OMNI v0.4.0 delivers intelligent update detection, a redesigned landing page, and critical Homebrew path stability fixes for seamless upgrades.
OMNI v0.1.1 delivers the first round of real-world bug fixes, tuning filter capture priorities under heavy concurrent IO shell streams.
OMNI v0.1.2 delivers critical micro-optimizations and edge-case bug fixes to stabilize the core CLI under heavy concurrent terminal loads.
OMNI v0.1.3 migrates to the Zig 0.15 IO API, replacing deprecated standard input/output functions with raw file stream equivalents for forward compatibility.
OMNI v0.2.0 replaces fragmented shell scripts with a unified, high-performance native CLI featuring distill, density, report, bench, and generate subcommands.
The very first release of OMNI proves that autonomous AI requires semantic filtering, not raw compression. Launching with Git and build log support.
Pre-release notes, kept as history. Not indexed, because nothing here is what finally shipped.