AI Build & Ask
Describe the change. Review the diff. Ship the app.
AI builds ObjectOS apps two ways: developers point a coding agent like Claude Code at the source project and it writes every metadata type, previewed in the browser and reviewed as a diff — or business users describe the change in conversation on Cloud. AI Ask answers questions and runs approved actions over live data, always inside the permissions of the person asking.
- 2 paths
- Build with a coding agent in source, or by conversation in the cloud
- ~1%
- Metadata surface the AI writes for CRUD/workflow apps
- 0 bypasses
- Every AI action runs inside user permissions and audit
The build loop
From one sentence to a governed application.
The loop is designed so speed never outruns control: the AI moves fast on the draft, and a person owns the decision.
- Describe: “Track vendor contracts with renewal reminders and an approval for anything over $50k.”
- Your AI — a coding agent in source, or the in-app builder — drafts the objects, fields, views, and renewal flow.
- Structural changes land in the approval queue as a compact, readable diff.
- You approve — the runtime ships tables, APIs, screens, automations, and audit.
- AI Ask now answers contract questions for each user, scoped to what they may see.
AI Build
Two ways to build — one reviewable definition
Whether the AI works in your repository or in a chat panel, the output is the same compact ObjectStack metadata a human reviews and owns.
In source, with your coding agent
Developers point Claude Code, Cursor, or any coding agent at the project; it writes every metadata type as source files, previewed in the browser and reviewed as a git diff — a Salesforce DX-style workflow.
In the cloud, by conversation
On Cloud and Enterprise, business users describe the change in chat; the in-app builder drafts it as metadata with live preview — Airtable-style ease with runtime governance.
Visual designers to fine-tune
Objects, views, flows, and dashboards each have a designer — drag-to-reorder fields, a flow canvas, kanban column config — and review mode diffs the draft against the published version, so tuning what the AI drafted never touches a file.
Approval before structure changes
Anything that changes the shape of the system queues for human sign-off with the full diff — on either path.
AI Ask
Answers and actions over live business data
AI Ask works inside the product, over real records — not a stale export — and it never sees more than the person asking.
Question the business
“Which contracts renew this quarter?” “Which vendors slipped SLA twice?” — answered from live objects with permissions applied.
Run approved actions
Ask can trigger actions that metadata explicitly exposes — assign, escalate, generate a summary — within the user’s rights.
Everything on the record
Every Ask query and action lands in the audit log alongside human activity, so review works the same for both.
Open source
Bring your own AI, keep the same governance
The open-source edition has no built-in assistant — by design. Your coding agent writes the same metadata as source files, and any MCP client queries the same governed objects.
Agent writes source files
Claude Code, Cursor, or any coding agent edits ObjectStack definitions in your repo; review happens as ordinary git diffs.
Any MCP client can ask
The MCP server exposes your objects, queries, and actions to whatever model you run — hosted or fully local.
Same runtime, same rules
Permissions, approvals, and audit apply identically. The edition changes who hosts the AI, never how it is governed.
Decision surface
What changes, who reviews it, what runs
| Edition | How AI builds | How AI answers |
|---|---|---|
| Community (open source) | Your coding agent edits metadata source files; review as git diffs | Any MCP client queries governed objects with your own model |
| Cloud Team & Business | In-app AI Build drafts changes; approvals gate structure | In-app AI Ask answers and runs approved actions |
| Enterprise | Same, plus private deployment and bring-your-own-model | Same, plus local models and internal identity |
FAQ
Questions this page should answer
What stops AI from breaking production?
Three runtime guarantees: structural changes require human approval, every action runs inside the requesting user’s permissions, and everything is audited. The AI can draft quickly precisely because it cannot ship silently.
Which models does it use?
Cloud editions run managed models behind AI Build and Ask. Open source brings your own — any MCP-compatible client or a local model. Enterprise adds bring-your-own-model for private deployments.
Next pages