Skip to main content

OpenWebUI Pipelines

Pipelines run alongside OpenWebUI to handle OpenAI-compatible traffic on port 9099. The Deployment mounts a PersistentVolumeClaim at /app/pipelines where pipeline files are downloaded on startup from the URLs specified in the PIPELINES_URLS environment variable.

Configuration​

  • Storage: A PersistentVolumeClaim (pipelines-pvc) with 5Gi of Longhorn storage is mounted at /app/pipelines to persist downloaded pipeline files across pod restarts. An emptyDir volume is mounted at /tmp for temporary files.
  • Secrets: The ExternalSecret app-openwebui-pipelines-api-key sources PIPELINES_API_KEY from Bitwarden and injects it into the container.
  • Service: The pipelines ClusterIP Service exposes port 9099 inside the open-webui namespace.
  • Access: OpenWebUI is automatically configured to connect to the pipelines service via the PIPELINES_URL and PIPELINES_API_KEY environment variables. The API key is sourced from the ExternalSecret app-openwebui-pipelines-api-key. Pipelines should appear automatically in the OpenWebUI admin panel.
  • Upstream provider: Configure pipeline valves in the OpenWebUI UI to call LiteLLM at http://litellm.litellm.svc.cluster.local:4000/v1 with the existing LiteLLM API key.

Available Pipelines​

The deployment includes several pipelines that are automatically loaded via the PIPELINES_URLS environment variable:

General Tools Pipeline​

Provides utility functions for LLM interactions:

  • get_user_name_and_email_and_id: Retrieves current user session information
  • get_current_time: Returns human-readable date and time
  • calculator: Evaluates mathematical equations (with security warnings)
  • get_current_weather: Fetches weather data from OpenWeather API

Configuration: Set OPENWEATHER_API_KEY valve parameter in OpenWebUI to enable weather functionality.

Document Search Pipeline​

Enables document retrieval from Qdrant vector store with hybrid search and RAG capabilities:

  • Supports semantic and keyword-based search
  • Optional reranking using Ollama or DeepInfra
  • Configurable result limits and filtering

Configuration: The pipeline defaults are pre-configured for cluster services:

  • Qdrant: http://qdrant.qdrant.svc.cluster.local:6333
  • Embeddings: nomic-embed-text:latest via LiteLLM
  • All settings adjustable via valve parameters in OpenWebUI

Wikipedia Pipeline​

Retrieves Wikipedia article summaries with automatic disambiguation and related topic suggestions.

Memory Filter​

Long-term memory filter using mem0 framework with Qdrant and Ollama for context-aware conversations.