# Legion

> Legion is an open-source, model-agnostic orchestration engine for AI agents. Give it a task and it plans the work, runs it across models and harnesses in parallel, cross-checks the result, and hands you one verified pull request.

Legion is the model-agnostic orchestration engine behind Opus Aether. It runs Claude, Codex, Cursor, opencode, and humans as one accountable workflow: plan, execute in parallel, cross-model verify, and stop at a reviewable pull request. Nothing is auto-merged. The core is Apache-2.0 and free forever. Enterprise support and managed pilots are available from Opus Aether.

The parts that make an AI agent stack work are not in a thread or a YouTube video — they are under the hood: the models, the harnesses they run in, the loops that verify and heal, and the self-improving agents built on top. Legion is exactly that layer — models, harnesses, loops, and self-improving agents — the stack, not another wrapper.

## What Legion does

- Plans a task by grilling the prompt across models, then decomposes the goal.
- Executes each slice on the cheapest capable model, in its own isolated git worktree, in parallel.
- Reviews by cross-checking across models and running your quality gates; nothing ships on one model's say-so.
- Ships a single pull request. Never auto-merged — a human approves every change.
- Watches itself: every step emits a metered `legion.span.v1` telemetry record (exportable as OpenTelemetry), and self-healing opens a fix PR when something breaks.

## Built for long-running agents

Legion was built to make overnight and continuous agents reliable — the overnight refactor, the nightly dependency sweep, the background watch on a build — where a single unbounded agent would drift. That is the difference between a coding assistant and an orchestration engine.

## Models and tools

- Routes current model roles across Claude Fable 5 and Opus 5, GPT-5.6 Terra/Sol/Luna, Cursor Grok 4.5, and opencode GPT-4.1.
- Runs across harnesses: Claude Code, Codex, Cursor, opencode, and Hermes, from one shared install.

## Data privacy

Legion runs locally on your machines with isolated git worktrees. It has no cloud of its own and no third-party vendor in the path (unlike tools that route your code through their cloud). Model inference uses your own Claude and Codex accounts, under your own provider terms.

## Install

- Quick start: curl -fsSL https://github.com/Opus-Aether-AI/legion-core/releases/latest/download/install.sh | bash
- Build on it as a package: npm i @opus-aether-ai/legion-core
- Source: https://github.com/Opus-Aether-AI/legion-core

## Pricing

- Open source: Apache-2.0, free forever. You bring your own model accounts; Legion adds no model cost.
- Enterprise: engagement-based, starting with a scoped paid pilot, then support and an SLA. See https://legion.opusaether.com/enterprise and https://legion.opusaether.com/pricing.md

## Links

- Site: https://legion.opusaether.com
- Enterprise: https://legion.opusaether.com/enterprise
- GitHub: https://github.com/Opus-Aether-AI/legion-core
- npm: https://www.npmjs.com/package/@opus-aether-ai/legion-core

## Contact

ai@opusaether.com