Ghost is the governed execution / trust runtime for AI agents — not another autonomous agent.
Agent (propose) → Ghost (approve · execute · verify · audit)
Teach Ghost a business workflow once. Agents (Claude, Cursor, Codex, …) may propose or start runs through Ghost’s tool surface; humans approve sensitive steps in Ghost; deterministic code executes; outcomes are verified; every run leaves a hash-chained audit log.
Capture → Review → Approve → Execute → Verify → Recover
Prefer, in order: APIs → browser automation → desktop → vision last.
Working name Ghost will likely be renamed (generic / hard to brand for trust). Strategy does not wait on naming.
Agents propose. Ghost runs the job with a human gate on anything sensitive, verifies the outcome, and leaves a record of exactly what changed.
Sell trusted execution into ops teams that already use AI assistants:
- wholesale / distribution order ops
- property management admin
- bookkeeping / month-end document flow
- logistics / recruiting / healthcare-admin back office
First money: implementation of one painful, measurable, reversible workflow
(+ seats). Distribution also rides agents via MCP / HTTP tools (see
cloud/docs/AGENT_PLUGIN.md).
- Not a chatbot or “AI coworker” that acts unsupervised
- Not competing on more autonomy than Claude/Cursor
- Not a Zapier/Make clone (connectors without a trust pipeline)
- Not a desktop Organizer utility as the commercial product
- Not vision-first RPA
| Surface | Status |
|---|---|
cloud/ SaaS + agent plugin API/MCP |
Active — build and sell this |
| Root Rust/Tauri desktop | Legacy — retained; not the roadmap |
cloud/README.md · cloud/docs/PHASE_1_PLAN.md · cloud/docs/CURSOR_HANDOFF.md ·
cloud/docs/AGENT_PLUGIN.md · AGENTS.md · this file · business-model.md