Skip to content

EmbrAIon in 60 Seconds

In plain English

You keep the rules for AI work with the repository. Your AI client reads generated host-native instructions, while EmbrAIon keeps project knowledge, safety boundaries, validation, and routing in one reusable project contract.

The setup path

pipx install embraion
  ↓
embraion init
  ↓
embraion install --host codex --destination .
  ↓
“Configure EmbrAIon for this project”
  ↓
project contract configured and verified
  ↓
ordinary engineering prompts

Use copilot or claude-code for other hosts. Project Bootstrap discovers knowledge, policy, and real checks; optional routing stays on host-default. Portable carries the capability but does not execute an AI session.

What you get

  1. Project knowledge — point the AI at the architecture and source-of-truth documents that matter.
  2. Project policy — classify canonical, protected, generated, and external paths.
  3. Validation — define real commands that prove a change works.
  4. Host projections — generate native agent/skill files understood by Codex, Copilot, or Claude Code.

Host projection = generated native files for your AI client. They deliver EmbrAIon instructions to the host; they are not a second source of truth. 5. Optional routing — classify work by complexity/risk and keep host-default model selection unless the project intentionally overrides it. 6. Optional enforcement — make protected-path, validation, and review requirements deterministic at delivery/merge time.

The normal experience

Configure the repository once, then work normally:

Add retry behavior when the connection drops.

You do not have to start every task with a framework prompt.

The smallest useful setup

For most repositories, focus on three things first:

What File
What should the AI know? .embraion/knowledge.yaml
What should it respect? .embraion/policy.yaml
How do we prove the change works? .embraion/validation.yaml

Everything else can stay at its safe default until you need it.

Two paths from here

  • Simple path: install → init → host projection → Project Bootstrap → normal AI tasks.
  • Engineer path: ownership → projections → routing → execution → evidence/enforcement.

One important distinction

Generated agent/skill files are instructions delivered to the AI host.

Commands such as:

embraion validation run affected
embraion enforcement check --base-ref origin/main

are real executable checks.

Next

Try it safely in the sandbox.