LLM connections

LLM connections are the brain of managed agents. You connect any number of providers, with the company's own API key, and choose which connection and model each agent uses. That way the company is not locked in to a single vendor.

Where to set it up

1. In the sidebar, open Integrations and choose the Company tab.
2. On the LLM providers card, click Configure.

Only owners, admins and people with the Manage AI agents permission see this area. See Roles and permissions.

Add a connection

1. Under LLM connections, click New connection.
2. Choose the Provider. The form fills in the right address for you.
3. Give the connection a Name you will recognise.
4. Check the API address and paste the API key.
5. Enter the Default model. Use Fetch models to ask the provider for the list of models available with that key.
6. Click Save connection.

The key is encrypted; once saved, only its last 4 characters are shown. To change it, paste a new one; leave the field blank to keep the current key.

Available providers

For Azure OpenAI and some compatible providers, adjust the Key header (Azure keys use api-key).

On hosted Team Connect, only public https addresses are accepted. On a self-hosted installation, you can also point to model servers on your own network.

API key, not a subscription login

Connections always use an API key from the company's account with the provider, billed under the company's commercial agreement. Personal or team subscription logins (such as Claude.ai or ChatGPT plans) cannot be used as an agent's brain, because those providers' terms do not allow it in third-party products.

Choose each agent's model

In the agent form, under Brain, choose the LLM connection and, if you like, a different Model. When blank, the agent uses the connection's Default model. This lets you, for example, use a cheaper model for reception and a stronger one for development. See AI agents.

Limit usage

On the same screen:

• Company monthly token limit: shows this month's usage and applies to every managed agent. Blank means no limit. When it is reached, agents pause until the month turns. Click Save limit.
• Keep transcripts for (days): how long what was sent to the model and its answers stay in the company database. Click Save retention.

Each agent can also have its own Monthly token limit.

Privacy and data sent (LGPD)

• The provider receives the task content, the agent's instructions, the company memories that apply to the task and, for coding agents, parts of the granted repositories.
• Team Connect does not train models or keep memory at the provider: what agents learn stays in company memory, in your own database.
• Keys never appear on screens, in logs or in the audit log. Creating, changing or deleting connections, limits and retention is recorded.
• Grant each agent only what it needs and avoid sending personal data the task does not require.