Kind: Server-side

The Model Source Call tool is the source-level counterpart to Model Endpoint Call. Instead of routing through an inference endpoint, it runs an inference pass directly against a single inference source. This is useful when you want a dedicated, low-latency inner call to a specific model instance without going through the endpoint's routing layer.

No Conduit agent is required - the call is executed entirely within the Infersec API.

How it works

  1. The model calls the tool with arguments matching the tool's input schema.
  2. Infersec sends a chat completion directly to the referenced inference source, applying the systemPrompt.
  3. The source's response (optionally validated against the outputSchema) is returned as the tool result.

Configuration

Field Required Description
description yes What the tool does; shown to the model
sourceId yes The target inference source to call
inputSchema no JSON schema describing the arguments the tool accepts
outputSchema no JSON schema describing the expected response
systemPrompt no System prompt applied to the inner inference call

The sourceId must reference an inference source in the same account. Because the call bypasses endpoint routing, the target source must be online when the tool is invoked.

Endpoint call vs source call

  • Use Model Endpoint Call when you want routing, fallback, and load distribution across multiple sources.
  • Use Model Source Call when you want to target one specific source directly.

Connecting

Server-side tools need no Conduit connection. Create the tool in the console, add it to a tool service, and attach that tool service to an inference endpoint.