Configuration reference

The worker is configured entirely through environment variables. This page lists the required deployment values and the analysis settings. For how to supply them during a deployment, see Deploy the analyzer worker.

Required variables

Set all seven of these for the worker to start and serve. A missing or malformed value raises at boot and crashes the worker, so the pod fails to start.

Variable	Description
INSTRUCTION_HUB_RUNTIME_BASE_URL	Base URL of the hosted Promptless Runtime the worker syncs with.
INSTRUCTION_HUB_INSTALL_TOKEN	The plih_ install token that authenticates this deployment to hosted Promptless.
INSTRUCTION_HUB_DEPLOYMENT_NAME	A human-readable name for this worker deployment.
INSTRUCTION_HUB_DEPLOYMENT_INSTANCE_ID	The stable instance identifier used when hosted Promptless verifies host credentials.
INSTRUCTION_HUB_CONFIG_HASH	A hash of the deployment configuration.
INSTRUCTION_HUB_CUSTOMER_POSTGRES_DSN	Connection string for the Postgres database you own.
INSTRUCTION_HUB_TRACE_OBJECT_S3_BUCKET	The S3 bucket you own, for raw traces and canonical trace objects.

Caution

Treat INSTRUCTION_HUB_INSTALL_TOKEN and INSTRUCTION_HUB_CUSTOMER_POSTGRES_DSN as secrets. Supply them through your secret store rather than as plain values.

Analysis variables

The INSTRUCTION_HUB_ANALYSIS_* variables turn on and configure the Friction Analyzer. Set them together, because the analyzer needs its model and repository configured before it runs.

Variable	Description
INSTRUCTION_HUB_ANALYSIS_ACTIVATION_AT	When analysis becomes active for this deployment. An ISO-8601, timezone-aware timestamp, for example 2026-07-29T00:00:00Z.
INSTRUCTION_HUB_ANALYSIS_MODEL_PROVIDER	The model provider: openai, azure_openai, or aws_bedrock.
INSTRUCTION_HUB_ANALYSIS_MODEL_AUTHENTICATION	How the worker authenticates to the provider: api_key or aws_sigv4. api_key works with any provider; aws_sigv4 requires the aws_bedrock provider and uses ambient AWS credentials, so it needs no API key.
INSTRUCTION_HUB_ANALYSIS_MODEL_BASE_URL	The provider’s base URL.
INSTRUCTION_HUB_ANALYSIS_MODEL_NAME	The analysis model to use. Defaults to gpt-5.6-sol.
INSTRUCTION_HUB_ANALYSIS_MODEL_API_KEY	The API key, when the provider uses API-key authentication.
INSTRUCTION_HUB_ANALYSIS_REPOSITORY_URL	The Instruction Hub repository the analyzer reads and remediates against.
INSTRUCTION_HUB_ANALYSIS_REPOSITORY_ID	The repository identifier.
INSTRUCTION_HUB_ANALYSIS_REPOSITORY_FULL_NAME	The repository’s full owner/name.
INSTRUCTION_HUB_ANALYSIS_REPOSITORY_TOKEN	A repository-scoped token the remediation flow uses to open pull requests. The analyzer also uses it to clone and read your Instruction Hub for its local mirror. Optional for public repositories.
INSTRUCTION_HUB_ANALYSIS_QUIET_WINDOW_HOURS	How long a session stays quiet before it is eligible for analysis. Defaults to 0.5.
INSTRUCTION_HUB_ANALYSIS_MIRROR_ROOT	Local root for the analyzer’s repository mirror. Defaults to /var/lib/instruction-hub/analysis-mirrors. The chart backs it with ephemeral pod-local storage (an emptyDir), not a persistent volume. The mirror is re-cloned on pod restart, and no PersistentVolume is required.

Caution

INSTRUCTION_HUB_ANALYSIS_MODEL_API_KEY and INSTRUCTION_HUB_ANALYSIS_REPOSITORY_TOKEN are credential material: the provider API key and the repository-scoped token the remediation flow uses. Supply them through your secret store rather than as plain values.

Observability variables

The worker also honors Datadog and Sentry environment variables to emit traces, logs, and errors. Those settings, the DD_* variables and SENTRY_DSN, are documented separately in Observability.