The open-source AI security harness

_{Backed by}

Investigate vulnerabilities, review findings, and work on fixes in your browser.
Use your own model provider. Connect coding agents through CLI or MCP.

Watch the 45-second harness launch film

https://github.com/user-attachments/assets/b0f0facd-a8bf-4642-91c7-31c0612e499e

_{The Swiss Applied AI & Cybersecurity Research Lab}

Get started

Use 0cloud for managed security reviews on your GitHub pull requests, or run 0 locally with your own model provider.

PR security reviews with 0cloud

Connect your GitHub repositories to 0cloud, the hosted service from 0.security. It checks pull requests with deterministic scanning and AI agents, runs targeted tests in a sandbox, and posts findings with evidence and suggested fixes. Review the findings and apply fixes from the same PR.

Start with 0cloud → · How it works · Pricing

Run 0 locally

Install on Apple Silicon macOS or x64/ARM64 Linux, then run 0:

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

0 opens the local browser. In Connections, add a model provider or supported subscription, then choose a workspace and start a chat. Local use needs no 0cloud account. Use the CLI or MCP for automation.

npm, Windows, source and Docker installation · Model connections · Troubleshooting

Findings

Review source evidence and organize findings in a queue or on the board.

Security workflows

Chat with 0 in the web UI, investigate findings, and use built-in templates for assessment, verification, fixes, and research. Runs retain results; Learning keeps local activity and source-grounded notes.

For automation and coding agents, the CLI exposes the same workflow runner:

0 workflow list --templates
0 workflow run --template security-research \
  --target /absolute/path/to/repo --workspace /absolute/path/to/repo
0 runs list
0 chat --prompt "your request"

Workflow guide · Learning · GitHub Actions

Hosts and plugins

Choose a host for your work. Connect service integrations and manage extensions. Use ⌘K / Ctrl+K to jump between tools.

Skills and memory

Reuse security instructions and retain lessons linked to the source.

Connect another agent

Use CLI commands or expose workflows through MCP:

0 mcp-server --workflows --workspace /absolute/path/to/repo

Local MCP requires an explicitly selected host-local execution profile. Browser chat's Connect an external agent provides a setup prompt. CLI and MCP workflow clients can share a running local engine's sessions and runs.

Plugins and MCP setup · Local and remote engines · Plugin development

For AI agents: product index · command reference.

Development

Install dependencies with pnpm install, run npm run dev, and open the printed browser address. The dashboard updates with hot reload. See CONTRIBUTING.md.

For local binary analysis, run npm run setup:0verse once. It finds Ghidra and Java, installs the Python analysis/MCP dependencies, and checks the decompiler. Source-checkout chats connect the local 0verse MCP server automatically; an explicit ZERO_MCP takes precedence, and ZERO_0VERSE_MCP=0 disables it. Use the capability report to check which analysis tools and executors are available. 0 binary ./target --mode scan --format ndjson uses the same runtime.

Status and safety

Assess authorized systems and review the evidence before applying fixes. Enable the scope plugin to enforce target boundaries. The default shell runs on your host; use an isolated executor for untrusted work.

Report vulnerabilities through SECURITY.md. Licensed under MIT OR Apache-2.0.

Public disclosures and upstream fixes.