# Living Capabilities: connectors, actions, and field mapping

**Summary:** Connect your CRM, billing, and product systems so the agent can act on the customer's behalf. Includes custom connectors and field-level mapping.

**In short:** One of the three senses. What the agent can do: agentic procedures that work in your CRM, billing, and product systems on the customer's behalf.

**Last updated:** 2026-08-10

## Key concepts

### Standard connectors
Pre-built connectors for common systems. CRMs (Salesforce, Hubspot). Helpdesks (Zendesk, Freshdesk, Jira Service Management, Zoho Desk, Xurrent). Knowledge bases (Confluence, Google Drive, BookStack, Zenya, ClickHelp, Azure DevOps Wiki, WordPress). Live chat (Zendesk Sunshine Conversations, Freshchat). Data lakes (Visma Data Lake). Configure with OAuth and field selection.

### MCP servers
Connect your own MCP server to expose internal tools, custom APIs, or proprietary data sources to the agent through a standard protocol. The agent reads and acts through MCP the same way it does through any other connector, with the same audit trail.

### Custom connectors
Build your own connector when a system is not on the standard list and MCP is not the right fit. Uses the Unless connector framework. Your developers, your account manager, or Unless Studio can scaffold one.

### Actions
Each connector can also expose actions: update a profile, change a subscription, create a ticket, trigger a renewal. You set the permission per action: the agent may run it alone, it needs a human to approve, or it is off limits. No irreversible action runs without a human approving the boundary first.

### Field mapping
Each connector exposes data to the agent through field mappings. You decide which fields the agent sees. Sensitive fields can be tokenized through the Privacy Vault.

### Refresh behavior
Connectors are read live at interaction time, not pre-synced. There is no cached copy of your CRM inside Unless. The agent queries the source when it needs the data.

### Failure handling
If a connector is slow or unavailable, the agent degrades gracefully. It still answers; it just cannot take the actions that connector would have made possible.

## What you can do here

- Connect a new standard connector
- Configure field mapping for an existing connector
- Set the permission for each action a connector exposes
- Build a custom connector (with developer or account manager help)
- Tokenize sensitive fields through the Privacy Vault
- Inspect the live data a connector exposes
- Configure failure behavior per connector

## When to use it

- Your system is not on the standard connector list
- You need the agent to see a field that is not in the default mapping
- You want the agent to do something in a system, not just read from it
- An existing connector is exposing too much (privacy review flagged it)
- You are setting up a new Moment that needs new data

## How it works

When the agent needs data or needs to act, it calls the connector. The connector reads from your system using your credentials, returns the requested fields, and the agent uses them in the conversation. PII passes through the Privacy Vault before it reaches any third-party generative model.

## Frequently asked questions

**How do I connect a new system?**
Open \

**How do I build a custom connector?**
Talk to your account manager. The connector framework can be scaffolded by the Unless team or by your developers using the documented API.

**How do I tokenize a sensitive field?**
In the field mapping, mark the field as sensitive. The Privacy Vault tokenizes it. The agent sees a token, not the raw value, except when answering directly to the data subject.

**Why is a connector returning no data?**
Three common causes. The user used to authenticate lost permissions. The field mapping points at a field that no longer exists. Or the source system is unreachable.

**How do I see what a connector exposes?**
Open the connector and click \

**How do I limit what a connector exposes?**
Edit the field mapping. Only fields on the mapping list are exposed to the agent. Everything else stays in your source system.
