# Guardian

> Guardian is a local-first desktop and CLI governance layer that enforces architecture, security, and release policies on AI-generated code changes with human approval built in.

Guardian is a desktop and CLI governance layer built for small engineering teams that need to control AI-generated code before it ships. Rather than functioning as a coding assistant or generic scanner, Guardian acts as the release decision layer for AI-assisted code changes, combining policy enforcement with human accountability.

The tool follows a four-step release governance cycle. First, it identifies and isolates AI-assisted or unusually large code changes for stricter review. Then it applies team-defined architecture, security, and quality policies automatically. Human reviewers validate findings and record approval or override decisions with documented reasons. Finally, Guardian produces a release readiness signal backed by evidence, answering whether the code can ship and why.

- **AI-Generated Code Intake** — Detects AI-heavy or large refactor pull requests and routes them into stricter evaluation paths before release.
- **Policy-as-Code Enforcement** — Applies architecture, security, and quality rules defined in your repository to every risky change, with strict, warn, or off gate modes.
- **Human Approval Workflow** — Requires named approvers and override reasons for high-risk flows, maintaining a durable audit trail of every release decision.
- **Release Decision Surface** — Produces explicit pass, pass-with-warning, or block decisions with supporting evidence rather than just listing issues.
- **Guru AI Assistant** — Explains risky findings in plain language and generates policy-aligned fix proposals using configurable AI providers including OpenAI and Ollama.
- **Semantic Code Search** — Finds similar code patterns across the codebase using a local vector database powered by sqlite-vec.
- **Reviews and Fix Proposals** — Provides a structured queue for reviewing, applying, or rejecting AI-suggested fixes with per-file undo history.
- **Local-First Architecture** — Policies stay in the repository and the desktop plus CLI workflow operates locally, with cloud AI providers optional.
- **CLI and CI Integration** — Shared scan policy core ensures consistent behavior between the desktop app, CLI, and CI release gates.
- **Multi-Language Localization** — Desktop and website UI support English and Turkish, with AI outputs following the selected language.

## Features
- AI-generated code change detection and intake
- Policy-as-code enforcement for architecture, security, and quality rules
- Human approval workflow with named approvers and override reasons
- Release decision surface with pass, warn, or block outcomes
- Guru AI assistant for explaining risky findings and suggesting fixes
- Semantic code search using local vector database
- Reviews queue for managing fix proposals with apply and undo
- Configurable scan scope with source, extended, and full modes
- CLI enforcement modes with strict, warn, and off gate behavior
- Audit trail for all release decisions and approvals
- Support for OpenAI, Ollama, and offline AI providers
- Web search integration via Tavily for contextual research
- Guardian.lock for reproducible team scans
- English and Turkish localization

## Integrations
OpenAI, Ollama, Tavily, GitHub Actions

## Platforms
MACOS, WINDOWS, CLI

## Pricing
Open Source

## Version
1.2.6

## Links
- Website: https://www.guardianide.com/en
- Documentation: https://www.guardianide.com/en/docs
- EveryDev.ai: https://www.everydev.ai/tools/guardian-ide