Skip to main content

BYOLLM (Bring Your Own LLM)

Bring Your Own LLM (BYOLLM) lets you register your own large language model endpoint -- in your own cloud account, with your own keys -- and use it for upcoming aprity scans instead of the models managed by aprity.

info

BYOLLM is available on the Intelligence plan only. It is not included in Documentation or in the Trial. See Plan Comparison.

European environment policy

The default European environment currently refuses customer-managed endpoints, including connection tests and registration. An endpoint needs separate residency qualification and explicit authorisation before BYOLLM can be enabled. Choosing an EU URL alone does not establish processing location or retention terms.

Why use BYOLLM​

  • Data control. The content sent to the model during a scan goes through your own provider account.
  • Cost control. LLM usage is billed by your provider, under your existing agreements.
  • Model choice. You choose the provider, the model and the deployment.

Supported providers​

  • Azure OpenAI
  • OpenAI
  • Anthropic
  • xAI

The configuration form offers three authentication modes: API key, Bearer token and Azure Entra ID. Use the one your provider and deployment accept.

Where to configure it​

BYOLLM is configured in Salesforce, in the aprity app: open the aprity tab, then BYOLLM in the Configure section of the side menu.

The BYOLLM entry is shown only:

  • to aprity administrators,
  • when your plan includes BYOLLM,
  • after a first scan has completed (the page uses it to estimate costs).

Register an LLM​

  1. Open the BYOLLM page and click Register LLM.
  2. Fill in the form:
    • Display name -- how the LLM appears in the list.
    • Provider and Auth mode.
    • Base URL, and if needed Request path, API version and API key header.
    • Model, and for Azure OpenAI the Deployment name.
    • API key (or token).
    • Input cost per 1M tokens and Output cost per 1M tokens -- used to estimate the cost of your next scan.
    • RPM limit and TPM limit -- the request and token rate limits of your deployment.
  3. Click Test connection. aprity sends a short test request to your endpoint.
  4. Click Save LLM. The endpoint is tested again before the key is stored; an endpoint that fails the test is not saved.

The Registered LLMs table lists each LLM with its provider, model, costs, the estimated cost of the next scan and its status. LLMs managed by aprity appear as Managed by aprity.

Choose the LLM for upcoming scans​

On the same page:

  1. Pick the Primary LLM for upcoming scans.
  2. Optionally pick an RPM/TPM fallback -- an LLM used when the primary one is throttled by its provider. Without a fallback, scans may take longer when the provider throttles requests.
  3. Save. The selection applies to the next scans; scans already running are not affected.

The page also shows the tokens consumed by the latest completed scan, to help you size your deployment.

Good to know​

  • Keep your key valid. If the key expires or is rotated, scans that call the model fail. Update the LLM with the new key.
  • Rate limits matter. A deployment with low RPM/TPM limits makes scans slower. Declare the real limits of your deployment.
  • Deployment policy. Your endpoint is treated as customer-managed. If your aprity environment does not accept customer-managed endpoints, saving or testing the LLM is refused with an explanation.

Frequently asked questions​

Can I switch providers later? Yes. Register another LLM and select it as primary. The change applies to the next scan.

Can I use a different LLM for each scan? The selection applies to all upcoming scans of your organization until you change it.

Does BYOLLM change the rest of the pipeline? No. Metadata extraction, dependency analysis and documentation rendering are unchanged; during a scan, the calls to the language model go to the LLM you selected.