Kritt-ai/open-kritt

What it is

open·kritt is a self-hosted platform (AGPL v3.0) from the Kritt/Blockian team that orchestrates AI agents to run parallelized, task-scoped vulnerability research over code, then de-duplicates and ranks findings. The summary covers its purpose; the actionable substance the page adds is around how it must be deployed safely and the details of running it.

Preconditions

• Requires Git, Docker with Docker Compose, and Node.js 20+.
• One model-access option is required: a Codex login, or an API key via OpenAI, Anthropic, DeepSeek, OpenRouter, or xAI. Agents run on Codex, Claude Code, or Grok Build.
• GITHUB_TOKEN is optional, needed only for private GitHub repositories.

What to do

Standard setup — no install step; the CLI is repository-local:

git clone https://github.com/Kritt-ai/open-kritt
cd open-kritt
./kritt setup
./kritt start

Then open http://localhost:5173. On a headless server, leave the stack running and use a second shell:

./kritt-headless

The headless CLI imports workflow/post-script/skill/ranker JSON, creates scans with the same backend validation as the web form, shows status and failure reasons, edits non-secret settings, and exports bundles — but does not display finding contents in the terminal.

Deployment-security constraints the page states explicitly:

• The backend has no application authentication. Default ports bind to 127.0.0.1; keep the stack private. Do not expose it.
• Tool-enabled agents run as root in disposable job containers with writable repo copies and direct internet access (to install tools, compile, test, build PoCs). Run on a dedicated Docker host or VM, and read the threat model before scanning untrusted code.
• In Accounts, the persisted Active switch controls which credentials may receive new assignments while keeping others saved.

Caveats

• Attacker-influenced report and PoC source is kept as plain text in exports — treat exported bundles as potentially hostile content, not trusted markup.
• Exports distinguish state: completed scans yield complete exports; stopped/failed scans with findings yield clearly marked partial exports. Each ZIP includes a "share-safe" manifest.
• The no-auth backend plus root-with-internet agents means a shared or network-reachable host is a direct compromise risk; the isolation requirement is not optional advice.