Copilot: Configuration

Copilot uses a Bring Your Own Model approach. To use the feature, your organization must provide access to a supported Large Language Model (LLM).

An administrator must configure Copilot so that users can access it.

Before you begin

Make sure you have:

• An account with a supported LLM vendor.
• Access to the model you want Copilot to use.
• Valid connection parameters to connect and authenticate requests to the LLM vendor (e.g. API keys, urls, etc. see tables below).

The availability and cost of a model depend on the selected LLM vendor.

Configure Copilot

To configure Copilot:

1. Open Admin & Settings.
2. Select Copilot Configuration.
3. Fill the configuration fields (it depends on the vendor, see the dedicated section below).
4. Select Save model definition.

After the configuration is saved, Copilot uses the selected model to process users' requests.

The Copilot Configuration is divided into three sections:

• Model definition: the LLM vendor, its credentials and the token limit of a request.
• Context definition: the business context sent to the model together with users' questions.
• Context limits: how much graph data is included in the context sent to the model.

Model definition

You first need to choose your LLM Vendor via the "Vendor" field. Supported vendors include Google, Anthropic, OpenAI, Mistral, Azure OpenAI and AWS Bedrock. We also support connecting to a LLM via Ollama (use the "Custom" vendor option). According to the vendor, different fields are available to connect to the LLM.

⚠️ API keys are sensitive credentials. Do not share them or include them in screenshots, support requests, or other documentation.

The Token limit field is available for all vendors, see Token limit.

Google or Mistral

OpenAI

Anthropic

Azure OpenAI

AWS Bedrock

Custom (Ollama)

Token limit

The optional Token limit field sets the maximum number of tokens Copilot can send to the model in a single request. Default value: 100 000. See how Copilot builds its context.

Context definition

The context definition is free text that gives the model additional context or instructions specific to your organization and your use cases, for example:

• how to formulate its responses
• which data to consider or to ignore (relevant node categories, edge types or properties, data that is unreliable or out of scope)
• steps to follow during an investigation

It is appended to the Copilot system prompt, so that Copilot can interpret users' questions and the graph data with your domain knowledge in mind.

ℹ️ The context definition is set per data-source: each data-source can have its own context, and only the context for the data-source the user is working on is sent to the model.

To set it:

1. Open Admin & Settings.
2. Select Copilot Configuration.
3. Open the Context definition section.
4. Enter your context and select Save context definition.

Context limits

When Copilot inspects what the user is currently looking at (a visualization or a case), the nodes and edges of that view are sent to the model. The Context limits configurations limit how much of that graph data is included.

Edges whose source or target node was not included are dropped as well, so that the graph sent to the model never contains orphaned edges.

⚠️ Raising these limits increases the number of tokens consumed by each request, and can cause the token limit to be reached faster.