Configuring the Request Auditor

Setting the LLM and model

For the Request Auditor, you can specify the LLM and model to use to review the sampled redaction requests.

If you do not configure a specific LLM and model for the Request Auditor, then Textual uses the LLM model set in LLM_DEFAULT_PROVIDER and LLM_DEFAULT_MODEL_NAME.

To configure the LLM provider for the Request Auditor, use the environment variable REQUEST_AUDITOR_LLM_PROVIDER. For the list of available providers, go to Configuring the default LLM provider and model.

To configure the model to use for the Request Auditor, use the environment variable REQUEST_AUDITOR_LLM_MODEL_NAME. For information about recommended models, go to Configuring connections to LLM providers.

Configuring request annotations and audits

To configure how Textual sends request content to the LLM and scores the results, set the following environment variables:

• REQUEST_AUDITOR_ANNOTATION_CHUNK_SIZE_TARGET - The target number of characters for content chunks. Used to split large request content before sending it to the LLM. The content is always split at the start of a new line. The default value is 5000, meaning that Textual sends around 5000 characters at a time to the LLM.
• REQUEST_AUDITOR_ANNOTATION_MAX_PARALLELISM - The maximum number of chunks to send to the LLM at the same time during NER annotation. The default value is 4.
• SOLAR_MAX_CONCURRENT_JOBS_AUDIT_REDACT_REQUEST - The maximum number of concurrent request auditing jobs that the Textual worker can run. The audit compares the LLM results to the results from the original request. The default value is 8.

Scheduling the removal of expired requests

Textual runs a scheduled service to remove expired requests.

To configure the number of hours between runs of the service, set the environment variable PURGE_RECORDED_REDACT_REQUESTS_INTERVAL_HOURS.

The default value is 1, meaning that the service to remove expired requests runs every hour.