# HELIX Engine – Antigravity Edition

> A local evidence, memory and deterministic execution layer for coding agents, tuned for Google Antigravity and Gemini.

HELIX Engine – Antigravity Edition is an open-source, MIT-licensed project by SuperHelix77 that adds a local evidence, memory and deterministic execution layer to coding agents. This edition is tuned for Google Antigravity (`agy`) and Gemini 3.8 Flash High. It re-architects the context management of the Codex-oriented HELIX 1.0 around Gemini's prompt-caching behavior.

## What It Is

HELIX sits between a coding agent and its tool output. It keeps exact tool results in a local encrypted evidence archive and returns compact receipts to the model instead. It exposes `goal`, `context` and `memory` tools over MCP and runs as a local loopback service. It can be started from the CLI, and an Electron desktop app shows runtime state, evidence, memory and troubleshooting.

## How the Antigravity Edition Manages Context

- **Lean Switch (`lean_run`)** condenses long pytest or git diff output into verified status receipts, keeps error tracebacks, and stores the full raw output locally for recovery on demand.
- **Continuous Context Control (`seal`)** archives completed task turns into encrypted stores and carries forward a compact handoff receipt (`helix.handoff.v1`).
- **Two-Tier Context Optimizer** combines scoped task retirement with intra-task aging, turning older tool results into compact references while preserving recovery handles.
- **`batch_inspect`** replaces many small file reads with one atomic multi-file range inspection.
- **Memory** keeps explicit project and task notes in local SQLite, with no LLM observer needed.

## Setup Path

HELIX requires Node.js 22.19+, npm and Git; Python 3 is optional for AST-aware Python indexing. After building from source, users symlink the CLI into the Antigravity binary directory and start the daemon. Antigravity then discovers the HELIX MCP tools and the `/helixcontext` skill. The same stdio MCP entrypoint can also serve Codex or other MCP clients.

## Benchmark Claims and Boundaries

The README reports a 50-task SWE-bench-style run comparing vanilla Gemini 3.8 Flash High with HELIX. It lists 10 solves for HELIX against 11 for vanilla, with billable tokens reduced by 61.95% and cached context by 86.42%. These are the author's own figures. The author states that a 95% savings figure is a target, not a result. HELIX runs authorized commands with the host account's permissions and is not a sandbox.

## Features
- Lean Switch compact receipts for command and test output
- Continuous Context Control with encrypted handoff sealing
- Two-Tier context retirement and aging
- Multi-file batch inspection
- Local encrypted evidence archive with on-demand raw recovery
- Deterministic local SQLite memory
- Goal engine and Tool Flow for bounded mechanical sequences
- Native Antigravity plugin and MCP server
- Electron monitor for runtime state, evidence and memory
- Optional read-only Claude-Mem bridge
- RTK output filtering integration

## Integrations
Google Antigravity, Gemini, MCP, Codex CLI, RTK, Claude-Mem

## Platforms
WINDOWS, MACOS, LINUX, API, CLI

## Pricing
Open Source

## Links
- Website: https://github.com/SuperHelix77/HELIX-1.0-Antigravity
- Documentation: https://github.com/SuperHelix77/HELIX-1.0-Antigravity/blob/main/docs/INSTALL.md
- Repository: https://github.com/SuperHelix77/HELIX-1.0-Antigravity
- EveryDev.ai: https://www.everydev.ai/tools/helix-engine-antigravity