# How to govern the apps and tools people build with AI > People across a company now build apps, agents and Slates with AI. Substrait puts every one in one registry, with an owner, a risk tier, a security check and an audit trail. URL: https://substrait.build/problems/govern-tools-built-with-ai/ Last updated: 2026-11-06 **The question:** People across our company are building apps and agents with AI. How do we see and govern all of them? Give every AI-built tool the same registry, the same rules and the same record. On Substrait, every tool, agent and page is registered with an owner and a risk tier, passes the organisation's policy before production, and leaves an audit trail nobody can alter. ## Why this is hard AI-built tools arrive from many people, through many AI models, at a pace no review board can match. A review of each change one by one is slow, and builders route around it. No review at all means nobody knows which tools exist, who owns them or what data they touch. ## What Substrait does **One registry for everything people build.** Substrait registers every tool as an object with an owner, a purpose, a risk tier and a version. Substrait governs five kinds: Slates, knowledge bases, skills, web apps and agents. **Rules set once, applied to every tool.** Each organisation sets its own policies and risk tiers in Substrait. A change that satisfies the policy is approved automatically. A change the policy flags goes to the right human approver, with deadlines and escalation. **Limits instead of reviews.** An approval on Substrait sets a boundary: which data, read or write, customer-facing or not. Builders change the tool freely inside that boundary. A change that crosses the boundary goes back through approval. **A record nobody can alter.** Substrait keeps an audit trail of every access and action. A platform admin cannot edit it, and it exports to your security monitoring. **A switch for every tool.** Every tool on Substrait can be rolled back to the last good version or switched off.
Substrait console: a builder's tools shown as cards, each with its status and environment.
Each builder's tools in one place. Product UI. Sample data.
## Questions and answers ### What is the difference between governing AI-built tools and governing data? Data governance catalogues what a company owns. Substrait governs what a company builds: each tool's owner, its risk tier, the data it was granted, the checks it passed and every change after that. ### Does governance slow builders down? Not on Substrait. An approval sets limits once: which data, read or write, customer-facing or not. Changes inside those limits ship without a new review. Only a change that crosses a limit goes back for approval. ### Can Substrait govern agents as well as apps? Yes. On Substrait an agent is a tool its builder owns. Each agent has its own identity, its own limits, a record of every action it takes for a person, and a switch that stops it. ## Related - [See every tool your people built](https://substrait.build/capabilities/tool-registry/): Substrait registers every tool, agent, Slate, skill and knowledge base with an owner, a purpose, a risk tier and a version, and gives every one a switch that stops it. - [Set the guardrails once, not for every change](https://substrait.build/capabilities/policies-and-approvals/): Each organisation sets its own policies and risk tiers in Substrait. Changes inside an approved boundary ship freely; only a change that crosses it goes back for approval. - [An audit trail nobody can alter](https://substrait.build/capabilities/audit-trail/): Substrait records every access and action in an audit trail that nobody, a platform admin included, can change after the fact. It is queryable, exportable to your security monitoring, and kept in storage you control. - [A security command centre, built in](https://substrait.build/capabilities/security-command-centre/): Substrait gives the security team one console for every finding across every tool, opening on one page that answers how exposed the company is right now, in the organisation's own severity names.