Respawn-style undo layer for AI agents

Gives autonomous AI agents a local-first, rollbackable undo button for their actions

Problem

As AI agents autonomously modify files and run commands, users lack a sense of safety and reversibility. Some are already experimenting with Rust-based, local-first agent undo tools, showing that 'what if the agent messes up' is a real pain point.

Solution

A local-first agent action log and undo layer: intercept and record side effects like file writes, shell commands, and API calls; each action type has a corresponding inverse; a timeline UI supports one-click rollback of a single step or an entire session; no cloud, all data stays local.

Users

Developers using coding or automation agents, security-sensitive teams, power users of AI tools

MVP

Intercept and log file writes and shell commands; generate inverse actions for common operations; timeline UI with single-step/batch undo; local SQLite storage; CLI and simple SDK for agents to integrate

Revenue

Open-source core + team subscription (audit logs, policy controls, multi-user); usage-based pricing per agent or operation volume

Why now

Agent autonomy is rising while safety mechanisms lag; before giants build this in, an indie dev can gain traction with a local-first undo layer