Permission memory layer for coding agents

Make coding agents remember your approvals so you stop clicking Yes all day

Problem

Users complain they have to babysit their terminal because the agent asks permission for every tool call; worse, telling it “always allow” doesn’t stick across harnesses, so they answer Yes a quadrigazillion times a day. It’s a high-frequency, interrupt-heavy friction that nearly everyone hits.

Solution

A local permission-policy proxy that sits between the agent and the terminal: it records each approval, turns it into reusable rules keyed by tool plus arguments, and auto-approves matching calls while still prompting for sensitive ones. Includes a visual rule panel with whitelists, path scopes, command prefixes, and one-click revoke.

Users

Heavy users of Claude Code, Codex, Pi and similar terminal coding agents

MVP

Intercept and log tool calls plus user decisions; generate editable rules from tool+arg prefixes; auto-approve on match with a log; TUI/menu to view, search, revoke rules; import/export rule files

Revenue

Open-source core plus paid Pro (team rule sync, audit log, cloud backup) or a one-time $19–39 license

Why now

Terminal agents are exploding while permission memory stays primitive—a constantly complained-about gap, fully local with no compliance burden