Local Agent Gate

An auditable execution gate for local coding agents so they can't silently delete or push

Problem

Agent tool calls — Claude Code running shell commands, writing files, calling MCP tools — mostly execute by default, and developers only find out afterward what happened. Subagent drift, accidental deletions and stray commands are increasingly common, while existing safety tooling is cloud-heavy and enterprise-oriented.

Solution

A locally running execution gate that intercepts agent tool calls, applies configurable rules (path allowlists, command denylists, read-only sensitive dirs) to allow, ask or deny, and keeps a full audit log. Deep integration with popular agent CLIs with zero cloud dependency.

Users

Developers and small teams letting coding agents operate directly on their local filesystem and terminal.

MVP

Intercept shell and file-write calls; simple path and command rules; allow/ask/deny actions; local audit log viewer; one-command Claude Code integration.

Revenue

Free open-source core rules; Pro subscription for team policy sync, audit export and Slack alerts.

Why now

Coding agents are gaining more local execution power while safety guardrails remain enterprise-grade, leaving a clear gap for individuals and small teams.