2026.16 Infrastructure Changes

3 min read

Changes on Application environment

Change	Environment Variable Name	Application Default Value (if env variable unset)	Example	Required	Applications	Short Description
Added	`FEATURE_FLAG_SAVE_MODEL_USAGE_DASHBOARD_UN_18889`	`false`	`true`	false	`next-chat` `next-admin`	Turns on cost dashboard view in the UI. The view surfaces the model usage and costs. Note: the feature is experimental. Docs: https://unique-ch.atlassian.net/wiki/x/IgBUfw
Added	`MODEL_COSTS_FILE`	`/var/unique/model-costs/default-costs.yaml`		false	`backend-service-chat`	Path to the YAML model cost sheet. Defaults to the default model cost configuration. Can be overridden. Replaces JSON-in-env pricing. Note: the feature is experimental. Docs: https://unique-ch.atlassian.net/wiki/x/SIAjWQ
Added	`CONFIGURATION_BACKEND_URL`	N/A	`http://configuration-backend.chat:8095`	`true`	all backend services	It's needed to be able to talk to `CONFIGURATION_BACKEND` for feature flags
Added	`FEATURE_FLAG_ENABLE_MCP_TOOL_REFRESH_UN_19255`	`false`	`true`	false	`web-app-admin`	Turns on or off the "Refresh Tool Settings" button in MCP hub "Tools" section
Removed	`FEATURE_FLAG_AGENTIC_TABLE_RERUN_BUTTON_UN_13901`				`web-app-chat`	This feature is now GA; feature flag not needed anymore
Removed	`FEATURE_FLAG_AGENTIC_TABLE_ROW_RERUN_UN_14903`				`web-app-chat`	This feature is now GA; feature flag not needed anymore
Changed	`FEATURE_FLAG_ENABLE_PDF_CONTENT_EXTRACTION`	`false`			`agentic-ingestion`	Default value changed from `true` to `false`.
Added	`FEATURE_FLAG_ENABLE_IMAGE_CONTENT_EXTRACTION_UN_17223`	`false`	`true`	`false`	`agentic-ingestion`	Enable/disable the image content extraction module. When `false`, the `/image-content-extraction` endpoint is not registered. Must be `true` to enable the feature.
Added	`AGENTIC_INGESTION_BASE_URL`	`""`	`"http://agentic-ingestion.python.svc.cluster.local:8081"`	`false`	`backend-service-ingestion-worker(-chat)`	Points to the `agentic-ingestion` service. Required to enable Image Content Extraction feature.
Added	`API_BASE`	-	`http://node-chat.finance-gpt.svc.cluster.local:8092/public`	`true`	`agentic-ingestion`	Base URL for the Unique AI API (node-chat). Used for all LLM vision completions.
Added	`REDIS_HOST`	`""`	`redis-service.chat.svc`	`true`	`agentic-ingestion`	Redis hostname for job queue and result storage (from Key Vault).
Added	`REDIS_PORT`	`10101`	`6379`	`false`	`agentic-ingestion`	Redis port (from Key Vault).
Added	`REDIS_PASSWORD`	-	`password`	`true`	`agentic-ingestion`	Redis password (from Key Vault).
Added	`REDIS_USERNAME`	-	`default`	`true`	`agentic-ingestion`	Redis username, if ACL is enabled (from Key Vault).
Added	`REDIS_USE_TLS`	`false`	`true`	`false`	`agentic-ingestion`	Enable TLS for Redis (auto-enabled if port is 6380).
Added	`REDIS_CLUSTER_MODE`	`false`	`false`	`false`	`agentic-ingestion`	Enable Redis cluster mode instead of standalone.
Added	`REDIS_CLUSTER_NODES`	`""`	`[{"host":"redis-0","port":6379}]`	`true` if Redis cluster mode	`agentic-ingestion`	JSON array of cluster nodes (required if cluster mode is on).
Added	`REDIS_CLUSTER_PASSWORD`	-	`password`	`true` if Redis cluster mode	`agentic-ingestion`	Redis cluster password (from Key Vault).
Added	`REDIS_CLUSTER_ENABLE_TLS`	`false`	`false`	`false`	`agentic-ingestion`	Enable TLS for Redis cluster connections.
Added	`REDIS_TLS_REJECT_UNAUTHORIZED`	`true`	`true`	`false`	`agentic-ingestion`	Reject unauthorized TLS certificates for Redis.
Added	`IMAGE_CONTENT_EXTRACTION_MAX_WORKERS`	`4`	`4`	`false`	`agentic-ingestion`	Max concurrent async task workers for image extraction jobs.
Added	`IMAGE_CONTENT_EXTRACTION_CHAT_COMPLETION_TIMEOUT`	`240000`	`240000`	`false`	`agentic-ingestion`	Timeout in ms for each LLM vision completion call.
Added	`IMAGE_CONTENT_EXTRACTION_REDIS_JOB_TTL_SECONDS`	`3600`	`3600`	`false`	`agentic-ingestion`	TTL in seconds for job results stored in Redis.
Added	`IMAGE_CONTENT_EXTRACTION_TIMEOUT`	`2.0`	`2.0`	`false`	`agentic-ingestion`	Timeout in seconds when polling for a task result internally.
Added	`LOG_LEVEL`	`INFO`	`INFO`	`false`	`agentic-ingestion`	Application log level.
Added	`MAX_CONTENT_LENGTH`	`67108864`	`67108864`	`false`	`agentic-ingestion`	Max request body size in bytes (64 MB).
Added	`CUSTOM_CA_CERT_PATH`		`/etc/ssl/certs/ca.pem`	`false`	`agentic-ingestion`	Path to custom CA certificate for outbound TLS and Redis CA.
Added	`AGENTIC_INGESTION_IMAGE_POLLING_DURATION_MS`	`3000`	`3000`	`false`	`backend-service-ingestion-worker(-chat)`	Polling interval in ms for async job status checks.
Added	`AGENTIC_INGESTION_IMAGE_TIMEOUT_MS`	`300000`	`300000`	`false`	`backend-service-ingestion-worker(-chat)`	Max wait time in ms for a single figure extraction job.
Added	`AGENTIC_INGESTION_IMAGE_MAX_RETRIES`	`3`	`3`	`false`	`backend-service-ingestion-worker(-chat)`	Retries per HTTP request to agentic-ingestion.
Added	`FEATURE_FLAG_ENABLE_IMAGE_CONTENT_EXTRACTION_UN_17223`	`false`	`true`	`false`	`web-app-knowledge-upload`	Feature flag to show Image Content Extraction configuration section in the folder ingestion settings UI.
Added	`IMAGE_CONTENT_EXTRACTION_LANGUAGE_MODELS`	`""`	`AZURE_GPT_4o_2024_0513:GPT-4o (2024-05-13)`	`false`	`web-app-knowledge-upload`	Comma-separated `MODEL_KEY:Label` pairs for the image content extraction language model UI dropdown. Must be set to enable the feature.

New required environment variable:

As part of this release, CONFIGURATION_BACKEND_URL is now a required environment variable for all backend services. This connects each service to the centralized feature flag backend (backend-service-configuration) for flag evaluation.

Without this variable set, services will fall back to env-only flag resolution, which does not support per-company targeting or rule-based evaluation. This can lead to inconsistent flag behavior and confusion around where flags should be configured.

Please ensure CONFIGURATION_BACKEND_URL is added to the environment variables of all backend services prior to rollout. The value should point to the backend-service-configuration

New Image Content Extraction feature:

Please refer to this documentation for detailed instruction about how to deploy the new agentic-ingestion service and enable the feature: Agentic Image Content Extraction (experimental)