Agents
Agents react to events in the platform and record a result for staff: a score on an application, a score and a potential on a person, or a verdict on an identity document. They run in the background, one run per event, and every run is kept for review. See Reviewing Agent Runs.
Agents are managed in Settings → Automation & AI → Agents. Staff need the automation.agents_management feature (see Permissions & Limits).
Agent Types
| Type | Runs on | AI capability | What it records |
|---|---|---|---|
| Application scorer | An application | Yes | A score and a justification for the application |
| Potential scorer | A person | Yes | A score and a justification for the person, plus a potential (high, medium or low) for each of their journeys |
| Document validator | An upload to a Passport/ID field | No | A verdict on the document and the details it proves |
The type of an agent is chosen when it is created and cannot be changed afterwards.
The Agents List
The list shows each agent's Name, Type, Trigger and Status. Agents you created come first, in alphabetical order; system agents come last, marked with a System badge.
| Status badge | Meaning |
|---|---|
| Enabled | The agent runs when one of its triggers fires |
| Disabled | The agent does not run. Runs already queued are skipped when their turn comes |
| Needs AI access | The agent is enabled and needs an AI model, but model access is off for the institution |
The menu on each row offers Edit, Runs and Change history. Edit is not offered for system agents, nor for AI-capable agents while AI model access is off. Change history lists every saved change to the agent: version, action, the fields changed, who made the change and when.
Creating an Agent
Select Create a new agent, fill in the form and select Create agent.
| Setting | Applies to | Behaviour |
|---|---|---|
| Name | All | Shown in the list, in run details and next to the agent's results in application lists |
| Type | All | Application scorer or Potential scorer. System agent types cannot be created |
| Application templates | Application scorer | The templates whose applications it scores. Leave empty to score applications of every template |
| Triggers | Scorers | The events that start a run (see below) |
| Model | Scorers | The model used, from the institution's active AI provider |
| Prompt | Scorers | Your scoring criteria, in plain language. Insert example fills in a starting prompt |
| Include attachments for score | Application scorer | Adds the text of the application's PDF attachments to what the model reads |
| Programmes and classes | Potential scorer | Limits the agent to people with a journey in what is ticked (also subjects and courses when the institution uses them) |
| Agent status | All | Enabled or Disabled. New agents start disabled |
Triggers
| Trigger | Application scorer | Potential scorer |
|---|---|---|
| Application is submitted | Yes | Yes |
| Application is updated | Yes | Yes |
| Application is started | Yes | |
| Profile is updated | Yes | |
| Person registers for an event | Yes | |
| Form is submitted | Yes |
The document validator has no triggers to pick: it runs each time a file is uploaded to a Passport/ID field.
If an event fires while a run for the same agent and the same record is still waiting in the queue, the event joins that run instead of starting a second one.
Application Scorer
The application scorer reads the application's fields (and, when included, the text of its PDF attachments) together with your prompt, and returns a Score and a Justification of two to three sentences. The scale is the one your prompt defines; when the prompt gives none, the score runs from 1 to 100, higher meaning a stronger fit.
Application scorers can also be managed from an application template's settings page, in its Agents section. Associate adds the template to the agent's templates; Disassociate removes it. Only application scorers are offered there.
Warning: An application scorer with no templates scores every application. Disassociating the last template from an agent therefore widens it to all templates rather than stopping it. Disable the agent to stop it.
Potential Scorer
The potential scorer reads the person's profile fields (without contact details or identity fields), their journeys, their 20 most recent applications, and the names and dates of the 20 most recent events they registered for and forms they sent. It returns a score from 1 to 100 with a justification. For each journey it covers, it may then suggest a new potential:
| Score | Potential the score implies |
|---|---|
| 70 and above | High |
| 50 to 69 | Medium |
| Below 50 | Low |
A suggestion may be one step away from what the score implies when the journey itself gives a reason, never two. Suggestions are made only for journeys whose potential would actually change.
Warning: The potential scorer applies the potentials it suggests itself, with no staff review. The agent form shows this warning when the agent is enabled. Make sure your DPO has approved this before enabling it.
When a suggestion cannot be applied (the person's profile fails validation, or the potential changed while the agent was running), it stays as a suggestion for staff to apply. See Potential suggestions.
Staff members and erased people are never scored. With programmes and classes ticked, a ticked programme or subject also covers classes and courses added to it later, and only those journeys and their applications are sent to the AI. Leave everything unticked to score everyone.
Document Validator
The document validator is a system agent: Full Fabric provides one per institution and keeps it up to date. It reads passports and ID cards on Full Fabric's own servers without an AI model, so it keeps working with AI switched off. Its verdicts and what it proves are described in Passport/ID Field.
AI Model Access
The application scorer and the potential scorer carry an AI capability: they call the institution's AI model. They need AI model access to be on in Settings → Automation & AI → Governance.
- While access is off, the type field shows "Requires AI model access. Enable it in Governance." and these agents cannot be edited.
- Turning access off disables every enabled AI-capable agent. Turning it back on does not re-enable them: an administrator enables each one again.
- Agents without the AI capability, such as the document validator, are not affected.
The Model field names the active provider. Providers are managed in Governance; see AI Console & Assistant.