duncatzat

duncatzat / vigils

Public

A local control plane for AI agents — see what they do, approve what matters, keep secrets out. Rust + Tauri + Chrome MV3.

vigils.ai agent-security ai-agents audit-log desktop llm
50
5
89% credibility
Found Jun 01, 2026 at 50 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

Vigils is a local-first control plane for AI agents. It sits between your AI assistant and the tools/data they touch, giving you visibility and control. Every tool call is recorded in a tamper-evident log, risky actions pause for your approval, and secrets are automatically hidden before they reach the AI or any logs. The project includes a desktop app for real-time monitoring, a CLI gateway, and a Chrome browser extension.

How It Works

1
🤖 You hear about AI agents doing risky things

You use tools like Claude Code, Cursor, or other AI assistants that can read files, hit APIs, and paste into web UIs on your behalf. You start wondering: what exactly did they do? What did they see?

2
🔒 You install Vigils to watch over your AI assistant

You download the desktop app or CLI tool for your computer. Vigils sits quietly in the background, watching everything your AI agent does without getting in the way.

3
👀 Every action gets recorded in a tamper-proof log

Every tool call, every file read, every API request is written to a private log on your computer. The log uses a special chain of math that makes any tampering obvious — like a receipt you can always trust.

4
🛡️ Risky actions pause and wait for your approval

When your AI assistant wants to do something sensitive — like deleting files, sending data to the internet, or using a secret key — Vigils stops and asks you first. You can approve it once, for this session, or block it entirely.

5
Your secrets never leave your machine
🖥️
Desktop app path

Open the desktop app to see a live feed of everything your agent did, review pending approvals, and browse through past sessions.

🌐
Browser extension path

Install the Chrome extension to automatically hide secrets before you paste or submit on AI websites like ChatGPT or Claude.

6
🔌 Your AI connects through Vigils instead of directly

You point your AI assistant at Vigils instead of its usual tools. Vigils checks each request against your rules, asks for approval when needed, and keeps a complete record — all without your data ever leaving your computer.

✅ You can see everything your AI did, and nothing leaked

You have a complete, tamper-proof record of every action. Sensitive operations required your approval. Your passwords and secrets were never exposed. You finally have visibility and control over your AI agents.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 50 to 50 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is vigils?

Vigils is a local control plane that sits between AI coding agents and the tools they call. When agents like Claude Code, Cursor, or Zed try to read files, hit APIs, or execute commands, Vigils intercepts those calls and gives you visibility and control. Everything stays on your machine: prompts, secrets, and the audit trail never leave your local environment. It combines a Rust-powered backend with a Tauri desktop app and a Chrome extension, acting as a firewall, approval queue, and tamper-evident ledger for agent activity.

Why is it gaining traction?

The pitch is simple: AI agents are powerful, but they're also a black box. Vigils solves the trust gap by making every tool call visible, reversible, and auditable. The SHA-256 hash-chained ledger means you can verify nothing was tampered with. The default-deny firewall means nothing runs unless you allow it. And the redaction engine strips secrets and PII before they reach logs or models. For teams worried about agents accidentally exposing credentials or making destructive changes, this is the guardrail that doesn't require trusting the agent's own safety features.

Who should use this?

Security-conscious development teams using AI coding assistants in regulated environments or with sensitive codebases. DevOps engineers who need audit trails for compliance. Individual developers who want visibility into what their agent is actually doing. It's less useful for casual users running low-risk scripts or teams already satisfied with their agent's built-in safeguards.

Verdict

Vigils addresses a real problem with a thoughtful architecture, but the 0.9% credibility score and 50 stars reflect an early-stage project. The documentation is thorough and the feature set is impressive for the size, but production readiness requires more community validation. Worth watching and experimenting with, but not yet a standard part of the AI development toolkit.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.