Install

Install Cerebe Software Factory
in three pieces.

A server-side critic on every PR, a local critic on every commit, and the matching agent context (CLAUDE.md / AGENTS.md / .claude/agents) wired into the repo. Each step is independently useful — install the gate without the agent context, or scaffold a product with the full set out of the box.

Install the GitHub App

Install the Cerebe GitHub App on a repo and every pull request runs through the multi-vendor adversarial critic (Cursor, Codex, Gemini, Grok, MiniMax) — a verdict lands as a check run. Calibrated prompt overlays are the hosted IP; every bypass is a logged, attributable audit event.

Set up the GitHub App →

Install the CLI + hooks

The local critic complements the GitHub App — it runs against every commit before the push, so most quality issues never reach CI. Subscription auth where possible (Cursor + Codex CLIs already logged in), API key fallback where not.

install bash

# Install the free-to-use cerebe CLI (+ cyclone) — checksum-verified binaries
curl -fsSL https://raw.githubusercontent.com/momentiq-ai/cerebe/main/install.sh | sh

cerebe --version   # verify

Scaffold config + wire hooks, then render the skill set:

setup bash

# One command: scaffolds cerebe/, detects your local AI CLIs
# (Cursor, Codex, …), writes config, and runs init to wire git hooks.
cerebe install

# Render the bundled skill + agent-context set declared in darkfactory.yaml:
cerebe skills install --all

cerebe install writes the critic portfolio to cerebe/config.json — tune it any time:

cerebe/config.json jsonc

// cerebe/config.json — the critic portfolio (written by 'cerebe install',
// then edit to taste). Local critics use your existing AI-CLI subscriptions.
{
  "version": 1,
  "critics": [
    { "id": "claude",  "name": "Claude",  "command": "claude", "enabled": true },
    { "id": "cursor",  "name": "Cursor",  "command": "cursor", "enabled": true },
    { "id": "codex",   "name": "Codex",   "command": "codex",  "enabled": true }
  ],
  "aggregation": { "policy": "min-complete-quorum",
                   "blockingSeverities": ["blocker", "high"], "quorum": 2 }
}

What cerebe init wired (pre-commit runs your build/test gate; post-commit runs cerebe review in the background; pre-push reads the per-SHA artifact and blocks on blockers):

.husky/post-commit + .husky/pre-push bash

# 'cerebe install' writes these hooks for you. What they do:
#
#   pre-commit   your project's build / lint / type-check (your quality gates)
#   post-commit  cerebe review — the multi-vendor critic runs in the background,
#                writing a per-SHA verdict under .git/agent-reviews/
#   pre-push     cerebe gate-push — blocks the push if a commit has open blockers
#
# Bypass a blocked push with a logged, attributable reason:
AGENT_REVIEW_BYPASS="reason" git push

Verify with `cerebe doctor`

Reports node version, hook paths, core.hooksPath, per-adapter auth (subscription or API key), and model availability. All [OK] means the local critic loop is wired correctly — your next commit will fire the post-commit critic and your next push will be gated on its verdict.

terminal bash

cerebe doctor

# Verifies the git hooks, detected AI CLIs, and config.
# All [OK] = the local critic loop is wired correctly.

04 — Optional

Get the agent context into your repo

The Cerebe Software Factory gate runs against your commits regardless of what AI agent (or human) authored them. To get the matching agent context — CLAUDE.md, AGENTS.md, GEMINI.md, .claude/agents/, and a cycle-doc planning skeleton — into your repo, pick the path that matches your starting point:

Spawning a new product? Use Cerebe Blueprint — the scaffold ships the full agent context out of the box. See the Blueprint guide to get started.

Retrofitting an existing repo? Use cerebe onboard — it analyses your repo and scaffolds the agent-context set against what's already there.

terminal bash

cerebe onboard --apply

The agent context is what teaches Claude Code, Cursor, Codex, and Gemini what "correct" looks like in your product. The critic gate enforces it; the agent context guides what gets written in the first place. Together they close the loop.

Four gates, layered

Pre-commit hook — your project's build / lint / type-check. Fast, project-specific, catches the obvious before commit lands.
Post-commit critic — multi-vendor LLM review runs in the background, writes a per-SHA verdict to .git/agent-reviews/.
Pre-push gate — reads the per-SHA verdict, blocks the push if any commit has unresolved blockers. Bypass via AGENT_REVIEW_BYPASS="reason" git push (logged).
GitHub App critic — server-side multi-vendor review on every PR. Authoritative gate, surfaces as a check run.

Cost shape

Subscription auth for Cursor + Codex consumes your existing seats — no per-call API-token billing where it works. Gemini + Grok use API keys at vendor rates. Server-side per-PR cost is itemized in the dashboard; see Cost Visibility.

Diff-aware routing skips trivial PRs; per-repo + per-critic cost visibility keeps spend legible.

Get Started

Need hosted CI + fleet dashboard?

The GitHub App covers the multi-vendor critic on every PR. The dashboard surfaces verdicts, per-critic findings, and per-PR cost across your fleet. SSO, SOC2 Type II evidence, BYOK, VPC deployment available.

See Enterprise →

Looking for the full adoption guide, schema reference, and policy recipes? Read the docs → · cerebe on GitHub →

Install Cerebe Software Factoryin three pieces.