Skip to content

Back to the catalog

0sec-labs/0
github.comrepo

0sec-labs/0

"0" is an open-source security workflow runner from 0sec Labs (YC-backed, with a Swiss Applied AI & Cybersecurity Research Lab) usable from a terminal, a browser "Command Center," or a coding agent via MCP. It ships 12 starter templates (e.g. a "security-research" template that produces a source review of a demo API with code locations and suggested fixes), supports scheduled/triggered runs with retained results and revision history, and integrates with GitHub, Elastic, Semgrep, Snyk, Linear, Jira, Cloudflare, Slack and Teams via plugins.

TL;DR Open-source CLI/browser/MCP toolkit for running AI-driven security workflows like automated source code review with suggested fixes.
Open original ↗
Document

What it is

0 from 0sec Labs is installed via a shell script and run through a shared workflow runner accessible from terminal, browser ("Command Center"), or MCP.

How it works

Install with:

curl -fsSL https://raw.githubusercontent.com/0sec-labs/0/main/install.sh | bash
export PATH="$HOME/.0/bin:$PATH"

Templates, workflows, and runs are distinct concepts: a template is a starting point, a workflow is a set of customized/reusable steps, and a run is one execution. Example invocation:

0 workflow run --template security-research \
  --target /absolute/path/to/repo --workspace /absolute/path/to/repo \
  --format json > run.json

A built-in "Learning" system retains local run activity, source-grounded notes, and an evaluated improvement history across runs; saved workflows keep immutable revisions with restore support (0 learning status). For agent integration, 0 mcp-server --workflows --workspace <path> exposes MCP tools to discover templates, create workflows, and manage runs; a local MCP server requires an explicitly selected host-local execution profile, while a remote/registered engine is reached via --backend production. Clients can also attach directly to an already-running engine's sessions:

0 mcp-server --workflows --engine-url http://127.0.0.1:3000 \
  --engine-token-env ENGINE_TOKEN --session SESSION_ID
0 sessions list --engine-url http://127.0.0.1:3000 --engine-token-env ENGINE_TOKEN

ENGINE_TOKEN must be set to the engine's bearer credential; attached clients share the browser's session/run lifecycle, and disconnecting leaves the engine running. The model provider used by 0 for workflows is configured separately from whatever agent session is connecting to it.

Preconditions / Affected

Native installer supports Apple Silicon macOS and x64/ARM64 Linux only. A model provider must be configured before AI workflows will run.

What to do

The terminal defaults to a "YOLO" mode; use 0 console --mode standard to require confirmations before actions. An optional scope plugin can enforce target boundaries (limiting what the tool is allowed to touch). The project explicitly frames itself as a research preview: assess only authorized systems and review findings/fixes rather than trusting them blindly. Vulnerabilities in the tool itself should be reported via SECURITY.md.

Caveats

It's self-described as a "research preview," and the default execution mode (YOLO, no confirmations) is a deliberate but risky default that operators should override for anything beyond throwaway/authorized testing.

Technologies named here

Related in AI Security