0security

The open-source AI security harness

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

Get started · Demo · Docs · Research · Releases · 0cloud

0 web UI: submit a demo API source review, follow the investigation, and inspect findings with suggested fixes

Backed by Y Combinator · The Swiss Applied AI & Cybersecurity Research Lab

## Get started Install on Apple Silicon macOS or x64/ARM64 Linux, then run `0`: ```bash 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 console without an account. `0 web` remains a compatibility alias. The interactive terminal UI is retired; headless CLI commands and MCP remain available. Configure a [model provider](https://docs.0.security/api-keys/) to run AI workflows. See [installation](https://docs.0.security/getting-started/) and [troubleshooting](https://docs.0.security/troubleshooting/). ## Findings Review source evidence and organize findings in a queue or on the board.

0 findings: review the queue, organize the board, and inspect source evidence

## 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.

0 workflows: choose a template, inspect the workflow graph, and edit a review step

For automation and coding agents, the CLI exposes the same workflow runner: ```bash 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](https://docs.0.security/workflow/) · [Learning](https://docs.0.security/learning/) · [GitHub Actions](https://docs.0.security/ci/github-action/) ## Hosts and plugins Choose a host for your work. Connect service integrations and manage extensions. Use ⌘K / Ctrl+K to jump between tools.

0 hosts and plugins: jump with keyboard search, choose a host, and browse integrations and extensions

## Skills and memory Reuse security instructions and retain lessons linked to the source.

0 skills and memory: browse built-in skills and inspect saved lessons with source references

## 0cloud The hosted version of 0, for managed security workflows on your team's repositories. Access is currently through the **[0cloud waitlist](https://0.security/waitlist/)**. Local use does not require a cloud account.

Orange and violet cloudscape from the 0.security website

## Connect another agent Use CLI commands or expose workflows through MCP: ```bash 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](https://docs.0.security/integrations/) · [Local and remote engines](https://docs.0.security/engine-connections/) · [Plugin development](https://docs.0.security/hackstore/) For AI agents: [product index](https://0.security/llms.txt) · [command reference](https://docs.0.security/commands/). ## 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](.github/CONTRIBUTING.md). For local binary analysis, run `npm run setup:0verse` once. It discovers a local Ghidra and Java installation, installs the Python analysis/MCP dependencies without removing existing engines, and verifies the decompiler. In this source checkout, web and terminal chats automatically connect the local 0verse MCP server. Nine tools cover scanning, triage, preflight, bug classes, findings, PoVs, reports, memory recall, and capability discovery. The capability catalog also lists research CLI commands with their MCP-supported or CLI-only status. Each chat has its own scan state. `0 binary ./target --mode scan --format ndjson` uses the same runtime. This does not install every platform executor or fuzzing environment. Inspect 0verse's capability report before requesting confirmation; unavailable analysis is not a clean scan. An explicit `ZERO_MCP` configuration takes precedence, and `ZERO_0VERSE_MCP=0` disables automatic connection. ## Status and safety **Research preview.** Assess only authorized systems and review findings and fixes. Use `--mode standard` with headless chat for explicit scope checks. The optional [scope plugin](https://docs.0.security/scope/) enforces target boundaries. Report vulnerabilities through [SECURITY.md](.github/SECURITY.md). Licensed under [MIT](LICENSE-MIT) OR [Apache-2.0](LICENSE). [Public disclosures and upstream fixes](https://0.security/research/#disclosures).