v8.4.0
This release adds Catalyst AI (CT Bot) — an opt-in chat widget that answers admin users' questions via read-only MCP tools — plus a fix to the MCP aggregate tool schema for models without numeric columns.
New Features
Catalyst AI (CT Bot)
A floating chat widget is available on every admin page for users whose CmRole has Catalyst AI enabled. Conversations stream replies over ActionCable, support attachments and clarifying-question cards, and record per-request token/cost usage as CtCostTracking. See the Catalyst AI feature guide for details.
Migration Steps
- Run the Add Bot Conversations generator, then migrate:
rails generate cm_admin:add_bot_conversations
rails db:migrate
The generated migrations create four tables — ct_bot_conversations, ct_bot_messages, ct_bot_artifacts, ct_cost_trackings — and add is_catalyst_ai_enabled (boolean, default false) to cm_roles.
The conversations migration adds a foreign key to users. If your user class maps to a different table (e.g. admins), edit the generated migration's foreign_key: target before running db:migrate.
- Register the new models in
config/initializers/zcm_admin.rb:
config.included_models = [..., CtBotConversation, CtBotMessage, CtCostTracking]
- Add your OpenRouter key to credentials (
rails credentials:edit):
catalyst_ai:
openrouter_api_key: sk-or-...
-
Ensure
config/cable.ymlexists — the assistant streams replies over ActionCable. Use the Redis adapter in production. -
Seed the Catalyst AI platform settings (
CmPlatformSetting.add_cm_default_settings) or create CT Bot Greeting, CT Bot Model, and CT Bot Queue via the Platform Settings page. -
Switch on Catalyst AI on each
CmRolethat should see the widget, and declaremcp_actionson the models the assistant may query.
Checklist
- Run
rails generate cm_admin:add_bot_conversations && rails db:migrate(adjust the user-table foreign key if needed). - Add
CtBotConversation,CtBotMessage,CtCostTrackingtoconfig.included_models. - Set the
catalyst_ai.openrouter_api_keycredential. - Confirm
config/cable.ymlis configured (Redis in production) and Sidekiq is running. - Seed/create the three Catalyst AI platform settings; restart the app and Sidekiq after changing them.
- Enable Catalyst AI on the relevant roles and add
mcp_actionsto assistant-visible models.