AI Agents & Console
Full Fabric's AI module brings an assistant directly into the platform so staff can work with their data using plain language. Instead of building filters, opening reports, or clicking through records, you ask a question and get an answer drawn from live institution data.
The module has three parts:
- AI Console — an always-available chat panel embedded in the staff interface. Staff ask questions, look things up, build quick charts, and make changes through conversation. See Using the AI Console.
- MCP server — the same tools, exposed through the open Model Context Protocol so staff can connect their own AI assistant (Claude, or any MCP-capable agent) to Full Fabric and work from the chat they already use. See Connecting Your Own AI Assistant.
- AI Agents — automated assistants that run in the background against defined triggers (for example, scoring an application when it is submitted). Agents reuse the same data tools as the Console but run without a person in the loop.
How it works
The assistant is powered by a large language model (Claude). On its own, a language model has no access to your institution's data — so Full Fabric gives it a set of tools: safe, permission-checked actions that read and write platform records. When you ask a question, the assistant decides which tools to call, runs them against your live data, and writes its answer from the results.
This means answers reflect what is actually in the system at that moment. The assistant is instructed never to invent records, IDs, or statistics — if it cannot find something, it says so.
What the assistant can do
Tools are grouped by the kind of record they work with. The assistant only ever sees the tools you are allowed to use.
| Area | Examples |
|---|---|
| Profiles | Look up people, search and filter, view statistics, create and update profiles, manage tags, add to or remove from a class, activate or suspend |
| Applications | Find applications, view details and attachments, see statistics, unsubmit |
| Application Templates | Browse templates, start an application from a template |
| Campaigns | List campaigns, view statistics, create and update |
| Events | Find events, view statistics, create |
| Forms | List forms, view submissions and statistics |
| Evaluations | Look up application evaluations |
| Programmes | Browse programmes and view statistics |
| Email Templates | Find shared email templates |
| Segments | Look up and list segments |
| Transcripts | List academic transcripts |
Note: Reading data (looking things up, statistics) happens immediately. Actions that change data — creating, updating, suspending, tagging — always show a preview and wait for your confirmation first. See Making changes.
Access and permissions
The AI Console is gated behind a dedicated permission and is only available to internal staff whose role grants it. Beyond that, every tool the assistant runs is checked against your existing permissions: the assistant can never see or change data you could not access yourself through the normal interface. If a request would exceed your permissions, the assistant relays the permission error rather than working around it.
Note: Because access is permission-checked on every request, two staff members can ask the same question and get different results, each scoped to what they are allowed to see.
Auditing
Every Console conversation and agent run is recorded for review, including the question asked, the tools that were called, and the response returned. This history can be reviewed in the platform's AI settings, with access granted by an administrator, to understand how the assistant is being used and to support data-governance requirements.