AI Tools and Actions

When you chat with the Admin AI Assistant, the model isn't guessing answers -- it has a set of tools it can call to read your data and perform actions. This article documents those tools so you understand exactly what the AI can do behind the scenes.

These tools are only available to the admin chat. The member-facing chat has no tools and is read-only.

The Six Admin Tools

Confirmation guarantee: Every tool that writes to the database returns a preview first and waits for your explicit "Create" / "Confirm" click. The AI cannot silently modify your data.

1. create_announcement

Drafts a new announcement and shows you a preview.

• Inputs: title (required), content (required), category (optional: general, urgent, event, ministry)
• Output: Always saved as is_active = false (draft). You publish it manually from the Announcements list.

2. create_event

Drafts a new event and shows you a preview.

• Inputs: title (required), starts_at (required, YYYY-MM-DD HH:MM), ends_at (optional), location (optional), description (optional)
• Output: Saved as inactive. You activate it manually from the Events list.

3. search_events

Finds events without modifying anything.

• Inputs: start_date, end_date, keyword (all optional)
• Output: Title, date, time, location for each match.

4. search_people

Finds people by name or membership status.

• Inputs: name, membership_status (member, regular_attender, visitor, inactive), is_active (boolean) -- all optional
• Output: Names and basic info, scoped to your permissions (team leaders only see their assigned groups).

5. update_person_status

Updates one person's membership status or active flag.

• Inputs: person_id (required), membership_status, is_active, reason (optional, e.g. "Moved out of state")
• Output: Shows current vs. proposed state, then waits for your confirmation. Admin permission required.

6. query_database -- the data DSL

This is the most powerful tool. It accepts a small JSON DSL describing what you want, and the AI converts your natural-language question into the right call.

Schema

{
  "entity": "people | events | groups | announcements | attendance",
  "operation": "count | average | sum | list | min | max | group_by",
  "field": "age | birthdate | created_at | ...",
  "filters": { "...": "..." },
  "group_by": "membership_status | gender | age_group",
  "limit": 20
}

Entities

• people -- congregation records (with anonymization applied)
• events -- calendar events
• groups -- groups and classes
• announcements -- announcement records
• attendance -- attendance entries

Operations

Filter Examples

The AI builds the filters object from your question. Common shapes:

• {"membership_status": "member"}
• {"is_active": true}
• {"gender": "female"}
• {"age_min": 18, "age_max": 30}
• {"birth_month": 2} -- February birthdays
• {"birth_day": 14} -- specific day of month
• {"created_after": "2026-01-01"}

Sample Translations

Audit Logging

Every tool execution is logged with:

• The tool name and arguments
• The acting user's ID
• The church ID
• Timestamp

Each tool request and each confirmed action is recorded in the system log so you can see exactly what the AI did and when.

Safety Recap

• No deletes. No tool can delete records.
• Drafts only. All create_* tools save as inactive.
• Preview + confirm. Every write requires your explicit confirmation click.
• Permission-scoped. Tools respect your role; team leaders cannot query data outside their groups.
• Read-only by default. Four of the six tools are pure reads.

Related Articles

• Admin AI Chat Assistant -- the user-facing experience
• What the AI Knows -- which fields and records each tool can see
• Troubleshooting AI -- when a tool returns an error