0.0
The project is in a healthy, maintained state
A self-hosted Rails admin with typed resources and a shared, audited operation boundary.
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
 Dependencies

Runtime

~> 0.3
>= 2.13
>= 8.1.4.1, < 8.2
 Project Readme

Rails AI Admin

Rails AI Admin is a self-hosted admin engine for Rails. Give your team an admin for the models you choose, then add a RubyLLM-powered assistant that can answer questions about records and prepare changes for review.

0.4.1 Public Preview · Documentation · Quickstart · RubyGems

A searchable list of articles with scopes, filters, CSV export and import

Edit records through typed forms Review a change the assistant proposes
The edit form for an article An assistant conversation with a change to confirm or dismiss

What you can do

  • Manage records. Configure forms, associations, searchable and filterable lists with saved views and bulk actions, soft delete with restore, dashboards, and CSV import and export for explicitly registered resources.
  • Keep access in your app. Use the host's authentication, tenant scoping, and policies to decide who can see records, fields, and actions. Track changes with record history and an audit log.
  • Ask questions and review changes in chat. The optional assistant searches records and calculates totals using the signed-in admin's permissions. When it proposes a change, for one record or up to 100 at once, a review shows the details before a person confirms it. Per-person token and cost limits and a retention period keep it in check, and an optional MCP endpoint exposes the same permission-checked tools. Its tests replay conversations recorded with real OpenAI, Anthropic and Google models through OpenRouter.

How it compares

Most Rails admin gems show your models first and add permissions on top. Rails AI Admin starts closed, and its optional assistant follows the same permissions and audited write path as the forms.

Gem Rails Default model exposure Authorization Audit and history AI chat and MCP License
Rails AI Admin 8.1 Registered only Required, default deny Built in Built in, optional and free MIT
ActiveAdmin 6.1+ Registered only Optional adapter Other gems None MIT
Avo 6.1+ Generated only Paid add-on Enterprise plan Paid add-ons LGPL-3.0 or commercial
Administrate 6.0–8.x Generated only Optional Other gems None MIT
RailsAdmin 6.0–8.x All models Optional adapter PaperTrail adapter None MIT
Madmin 7.0+ Generated only Host application Other gems None MIT
Trestle 6.0+ Registered only Host application Other gems None LGPL-3.0

Choose another gem if you need Rails 7 or 8.0, an established plugin ecosystem, or an admin for every model without connecting sign-in and policies first. The full comparison has sources and more detail. Data as of November 2026.

Install in a Rails app

Add the gem to your Gemfile and run bundle install:

gem "rails_ai_admin", "~> 0.4.1"

Requires Ruby 3.3+ and Rails 8.1.4.1+ (below 8.2). Rails 7 and 8.0 are not supported. See the installation guide for supported Rails patch versions. The quickstart walks through connecting sign-in and policies, then registering your first resource. Conventional admin pages work without a model provider or API key. If you already use the engine, follow the upgrade guide to copy and run the new migration.

For an app with Rails authentication or Devise already migrated:

bin/rails generate rails_ai_admin:install --full-access
bin/rails db:migrate
bin/rails generate rails_ai_admin:resource --all
bin/rails rails_ai_admin:doctor

Review the generated files before use. The opt-in full-access preset grants all rows and declared capabilities to people admitted by AdminContext; add tenant scoping for account-based apps. Omit --full-access to keep policy access closed. The assistant remains off until explicitly enabled.

Try the local demo

From a clone of this repository:

bundle install
bin/demo --assistant

Open http://127.0.0.1:3917/login, choose Sign in as Admin, then select Assistant → New conversation. Try search articles for upgrading, followed by set articles 1 title to Hello to see a change review. The demo uses seeded data and an offline model; bin/demo --openrouter uses real models with an OPENROUTER_API_KEY from your environment. Follow the demo walkthrough for more examples.

MIT License · Contributing · Security