I built AgentLeash because I got uncomfortable.
I use Claude Code daily. It's genuinely useful. But one day I realized I had no idea which files it had actually read during a session. It could have accessed my .env, browsed through old projects, read things I'd forgotten were there. I had no visibility.
That felt wrong.
So I built something simple: a way to watch what AI agents do with your file system, and a way to tell them where they can and can't go.
What it does
AgentLeash monitors file operations and lets you define boundaries.
You create a .agentleash.yml file in your project:
default_policy: deny
rules:
- path: "src/**"
allow: [read, write]
- path: ".env*"
deny: [read, write, delete]
reason: "Contains secrets"
- path: "node_modules/**"
allow: [read]
Then you run leash watch and see everything in real-time:
14:32:01 [✓] READ src/components/Button.tsx
14:32:02 [✓] WRITE src/components/Button.tsx
14:32:05 [X] READ .env.local [BLOCKED]
That's it. You know what happened. You control what's allowed.
Key features
- Real-time monitoring of file operations
- YAML-based rule configuration
- Glob pattern matching (src/*, .config.js, etc.)
- Agent-specific rules (different rules for Claude vs Cursor)
- Violation alerts
- Session logs you can review later
Who it's for
Anyone using AI coding agents who wants visibility into what they're actually doing. Works with Claude Code, Cursor, GitHub Copilot, or anything else that touches your file system.
Pricing
- Free: 1 scope, 1,000 logs/day
- Pro ($19/mo): 5 scopes, 10,000 logs/day, custom rules, export
- Team ($49/mo): 20 scopes, webhooks, team sharing
A note on why I build these tools
AgentLeash is part of VERIDIAN, a small collection of developer tools I'm building solo. 15% of all revenue goes to funding free, emotional tech sanctuaries hosted by my MIRROFALL engine rolling out next month. No dark patterns, no tracking, no manipulative pricing.
I'm not trying to build the next unicorn. I'm trying to build honest software that solves real problems and funds something good in the process.
If AgentLeash is useful to you, that's enough.
Links
