Skip to main content

How to Bring Your Own LLM

Bring Your Own LLM (BYOLLM) lets your analyses use a model endpoint that your organization contracts and pays for, instead of the models managed by aprity. You register the endpoint yourself in the aprity app.

Prerequisites​

  • Intelligence plan. BYOLLM is not included on Documentation or Trial.
  • The aprity Admin permission set.
  • At least one completed analysis: until then the page shows Run a first scan before configuring BYOLLM.
  • An endpoint and credentials from a supported provider: Azure OpenAI, OpenAI, Anthropic or xAI.
  • A model with a context window of 1 million tokens: aprity is designed for this context size.

Register your LLM​

  1. Open the aprity app in Salesforce and click BYOLLM in the left navigation (section Configure).
  2. Under Registered LLMs, click Register LLM.
  3. Fill in the form: display name, provider, authentication mode (API key, Bearer token or Azure Entra ID), base URL, request path and API version if your provider needs them, API key header, Model and, for Azure OpenAI, Deployment, then the API key.
  4. Optionally, enter the input and output costs per million tokens (used to estimate the cost of an analysis) and the RPM / TPM limits of your endpoint.
  5. Click Test connection and wait for Connection verified.
  6. Click Save LLM.

The Registered LLMs table lists your endpoints next to aprity's standard model (its estimate reads Managed by aprity), with their costs, an estimated cost for the next analysis and their last test status. Click the edit icon on a row to change one of your endpoints.

Choose the model for your analyses​

  1. In Primary LLM for upcoming scans, select the model to use.
  2. Optionally, select a second model in RPM/TPM fallback. It takes over when the primary provider throttles requests; without one, analyses may take longer under throttling.

The choice applies to the next analyses, not to an analysis already running. The top of the page recalls the latest completed analysis and its input and output tokens, to help you size your endpoint.

Good to know​

  • BYOLLM affects the steps where AI writes or explains. Metadata extraction, dependency analysis and the other deterministic steps do not use an LLM.
  • Token usage on your endpoint is billed by your provider.
  • The documentation quality depends on the model you choose.
  • To go back to aprity's models, select aprity's standard model as primary.