Skip to main content
Pipelines are a plugin system that injects custom logic into the LLM request/response processing pipeline. Deploy Python code (.py) to the pipeline server to implement message pre/post-processing, custom models, etc. Configure in Admin > Settings > Pipelines.

​
What is a Pipeline?

Pipelines come in two types.
TypeRoleShown in Model List
FilterRequest pre-processing (inlet) + response post-processing (outlet)✗
PipeBehaves like a custom model✓

​
Filter Pipeline Flow

  • inlet: Process the message before passing to the LLM (e.g., prompt transformation, data injection)
  • outlet: Process the LLM response before delivering to the user (e.g., format conversion, filtering)
  • With multiple filters, they run in priority order
  • pipelines: ["*"] applies to all models; specific model ID list applies only to those models

​
Connecting a Pipeline Server

Pipeline server URLs aren’t entered separately in this tab. Servers added as OpenAI-compatible servers in the Connections tab that provide pipelines are auto-detected and shown in the dropdown.
To use pipeline servers, first register the server as an OpenAI-compatible API in Admin > Settings > Connections.

​
Installing a Pipeline

1

Pick pipeline server

Pick a pipeline server from the dropdown.
2

Add pipeline

Choose between two methods:
MethodDescription
GitHub URLEnter a GitHub Raw URL to download directly on the server
File uploadUpload a local .py file directly
3

Configure Valves

Selecting an installed pipeline shows the configuration parameter (Valves) edit form for that pipeline.
Pipelines are plugins that execute arbitrary code on the server. Don’t install pipelines from untrusted sources.

​
Valves Configuration

Each pipeline has its own settings (Valves) defined in code. Selecting a pipeline auto-generates the corresponding settings form.
Valve TypeUI Display
BooleanToggle switch
Enum (choices)Dropdown
Array (list)Comma-separated text input
Others (string, number)Text input
Each Valve is controlled by a None (use default) or Custom (set manually) toggle.
Valve changes take effect immediately on save. For Pipe-type pipelines, the model list is also auto-refreshed.

Connections

Register pipeline servers as OpenAI-compatible APIs

Tool Servers

Configure tool servers that agents call as external APIs