Skip to main content

Connect Your Data

An agent that can only talk is limited. An agent that can check a policy, a claim or a loan balance is useful. To do that, it needs a connection to the system that holds the data. This page explains what that involves, what to ask your IT team for, and how your credentials stay protected.

The idea in one minute​

Think of a connection as a saved, named link to one of your systems. You set it up once. Agents then use it to look things up. Achilles calls this a datasource.

You want the agent to reachTypical connection
A policy, claims or loan databaseDatabase connection
A web service such as a CRM or payments systemAPI connection
Email or other tools your team already usesTool connection, set up under Integrations in the console

You may also see the term MCP. It is a standard way of plugging existing tools into an agent. Your IT team will know it. As a business user you simply pick the tools you want from a list.

What to ask IT for​

Send your IT team this checklist:

  1. The name and type of each system the agent should use.
  2. A read-only account for each one. See below.
  3. The connection details for that account, shared securely, not by email or chat.
  4. Confirmation that Achilles is allowed to reach the system across your network, for example through an approved IP address.
  5. Which tables or services the agent should be allowed to see. Fewer is better.
  6. A test record you can safely look up.

What "read-only" means​

A read-only account can look at data but cannot change or delete it. It is the safest way to start. Even if something went wrong, the agent could not alter a policy or a balance.

When you later want the agent to take actions, such as updating a contact number, IT can grant a narrow, separate permission for that action. Actions also need approval. See Approvals.

How your credentials are kept safe​

  • You enter connection secrets once, in the console under Datasources. They are stored encrypted.
  • Secrets are never shown again after saving, and never appear in the agent's instructions, in test conversations or in exports.
  • The AI never sees your password. The platform makes the connection, not the model.
  • Changes take effect immediately, with no restart.
warning

Never paste passwords or connection details into an SOP, a chat, or an agent's instructions. Use only the Datasources page.

How to add a datasource​

  1. Open Datasources in the console menu.
  2. Choose Add datasource and give it a clear name, such as "Policy database".
  3. Pick the type and paste the connection details from IT.
  4. Choose Test connection. You should see a success message.
  5. Save.
  6. Go to Tools and choose which lookups use this connection. See Agent abilities.

If the test fails, check with IT that the account works and the network allows the connection. See Troubleshooting.

Protect what you share​

Even with read-only access, be selective.

  • Expose only the data the agent needs.
  • Sensitive fields such as ID numbers can be hidden from the AI. See Personal data.
  • Different people should see different records. See Roles and access.
note

For the technical side of integration, share Integration overview with your IT team.