[data-reveal]{opacity:1!important;transform:none!important}
AntelopeJS DMS
Star 25 GitHub repository (opens in a new tab)
All modules

Module 01 / 10 · AI assistant

Build and refine your admin

Tell the dashboard what to do.
Stay in control.

Create a page, reshape a layout or add a chart with an assistant that already knows DMS conventions. Work from a natural-language request instead of explaining your schemas, blocks and permissions from scratch.

Read the docs
npx @antelopejs/core project modules add @antelopejs/dms-ai
package
@antelopejs/dms-ai
status
Live
license
Apache 2.0

02 What it does

What the AI assistant module adds

Create and refine pages, layouts and configurations with an assistant that understands the DMS.

  • DMS-aware skills. The assistant understands block types, schema rules, layouts and role conventions. Add your own skills for the workflows specific to your product.
  • Safe mode. The assistant changes your project only through the Builder's tools: no direct file edits or shell commands, and every change must compile.
  • Vibe mode. Let the assistant write custom code when a workflow needs more than the standard components.
  • Reviewable changes. Every change the assistant makes is a diff in your repository, ready to review and revert with Git.

Permissions The AI never gets more access than the user it works for. Roles go down to each block, for people and for the assistant alike.

03 Two modes

It builds in development. Soon, it operates in production.

The generated back office is a starting point, not a black box. Whether you change it with the AI, in the builder or in your editor, every page stays clean, typed code in your repository.

In development

It builds.

Describe a change in the chat. The assistant works through the builder, so you review its work.

  • Build it visually. Click Edit this page. Each change rewrites the TypeScript source and is saved only if it compiles.
  • Safe mode. The assistant changes your project only through the Builder's tools: no direct file edits or shell commands, and every change must compile.
  • Reviewable changes. Every change the assistant makes is a diff in your repository, ready to review and revert with Git.

In production

It operates, as the user.

Next for the AI module: in production, the assistant works for your whole team. It sees and does what the connected user can, minus a global exclude list you control, and every action it takes is logged.

  • Ask your data. Support, sales and ops ask in plain words and get answers from live data, limited to what their role can open.
  • Scheduled prompts. Turn a prompt into a cron: a weekly churn report, a nightly check. It runs with its creator's permissions and can act on the dashboard.
  • An MCP server for your agents. Once enabled, every action your users can take in the dashboard is exposed over MCP. Plug in Claude Code or any agent with a user's token.
  • Documents on demand. Reports and documents generated when you ask, or on a schedule.

04 Guardrails · Your data stays yours

An AI on your production database? Only with guardrails.

The rules are built into the DMS, not left to a prompt. The AI never gets more access than the user it works for.

AI Builder Live

  • It builds from the schema, not your rows To map your database, the AI uses the catalog and aggregates only. Credentials stay in an environment variable.

  • Read-only until you decide Generated pages don't write by default, secret columns are left out, and nothing runs before you press Run.

  • Scoped like a teammate Roles go down to each block, for people and for the assistant alike.

AI Operator Coming soon

  • As the user, never more It sees and does what the connected user can, checked on every call.

  • A global exclude list Tables, columns and actions no AI, cron or connected agent can reach, whatever the role.

  • Crons run as their creator With the creator's current permissions, checked at each run.

  • MCP is off until you enable it An external agent connects with a user's token and gets that user's access.

  • Every action is logged Attributed to the user it ran for.

Every guardrail