OperateIQ
One AI operating platform for SMB industrials — where the business and its AI agents work off the exact same data, permissions, and audit trail.
Overview
SMB distributors and service companies run on 27 spreadsheets and no real software. OperateIQ replaces all of it with one platform — work orders, dispatch, inventory, procurement, invoicing — built on an idea I haven't seen elsewhere: every entity is exposed twice, once to a human through six universal renderers and once to an AI agent through ~255 shared tool definitions. Same service layer, same row-level security, same approval policy, same audit log. Your copilot can't do anything you couldn't, and it leaves the same paper trail.
A Kernel/Trait/Overlay model lets every acquired operating company keep its own field names while rolling up clean, so a new vertical ships as data, not code. I threw away a working 167-table v1 to rebuild on the cleaner model — conviction over sunk cost. Then I ran a six-month simulated operation across three orgs, driven entirely through the agent API, to stress-test the thing before a single customer touched it.
Highlights
- One service layer, two interfaces: humans (tRPC) and agents (MCP/REST) inherit identical permissions, policy, and audit.
- ~150 Postgres tables, each with tenant isolation, versioning, soft-deletes, and an event outbox on every write.
- Tiered model router — a cheap model for lookups, a mid one for natural-language-to-SQL, the top one for multi-entity reasoning.
- ~208K lines of TypeScript, built solo.