6f8c871658
- Fixed shared client pool bug to eliminate TTFT latency. - Added pure BYOK-only mode support. - Improved cross-user environment isolation for concurrency. - Resolved RichUI infinite vertical sizing loop. - Integrated client.ping() into stall detection. - Automatically hide TODO List widget after completion. - Synced all documentation and release notes.
3.6 KiB
3.6 KiB
Pipe Plugins
Pipe plugins create custom model integrations or transform LLM responses. They appear as selectable models in the OpenWebUI interface.
What are Pipes?
Pipes allow you to:
- :material-api: Connect to external AI APIs (Gemini, Claude, etc.)
- :material-robot: Create custom model wrappers
- :material-cog-transfer: Transform requests and responses
- :material-middleware: Implement middleware logic
Available Pipe Plugins
- GitHub Copilot SDK (v0.11.0) - Official GitHub Copilot SDK integration. Features Workspace Isolation, Zero-config OpenWebUI Tool Bridge, BYOK support, and dynamic MCP discovery. NEW in v0.11.0: Shared Client Pool fix (High Performance), Pure BYOK-only Mode, and Stable RichUI auto-sizing. View Deep Dive | View Advanced Tutorial | View Detailed Usage Guide.
- Case Study: GitHub 100 Star Growth Analysis - Learn how to use the GitHub Copilot SDK Pipe with Minimax 2.1 to automatically analyze CSV data and generate project growth reports.
- Case Study: High-Quality Video to GIF Conversion - See how the model uses system-level FFmpeg to accelerate, scale, and optimize colors for screen recordings.
How Pipes Work
graph LR
A[User selects Pipe as Model] --> B[Pipe receives request]
B --> C[Transform/Route request]
C --> D[External API / Custom Logic]
D --> E[Return response]
E --> F[Display to User]
The pipes Method
Defines what models this pipe provides:
def pipes(self):
return [
{"id": "my-model", "name": "My Custom Model"},
{"id": "my-model-fast", "name": "My Custom Model (Fast)"}
]
The pipe Method
Handles the actual request processing:
def pipe(self, body: dict) -> Generator:
# Process the request
messages = body.get("messages", [])
# Call external API or custom logic
response = call_external_api(messages)
# Return response (can be streaming)
return response
Quick Installation
- Download the desired pipe
.pyfile - Navigate to Admin Panel → Settings → Functions
- Upload the file and configure API keys
- The pipe will appear as a selectable model
Development Template
"""
title: My Custom Pipe
author: Your Name
version: 1.0.0
description: Description of your pipe plugin
"""
from pydantic import BaseModel, Field
from typing import Generator, Iterator, Union
class Pipe:
class Valves(BaseModel):
API_KEY: str = Field(
default="",
description="API key for the external service"
)
API_URL: str = Field(
default="https://api.example.com",
description="API endpoint URL"
)
def __init__(self):
self.valves = self.Valves()
def pipes(self) -> list[dict]:
"""Define available models."""
return [
{"id": "my-model", "name": "My Custom Model"},
]
def pipe(
self,
body: dict
) -> Union[str, Generator, Iterator]:
"""Process the request and return response."""
messages = body.get("messages", [])
model = body.get("model", "")
# Your logic here
# Can return:
# - str: Single response
# - Generator/Iterator: Streaming response
return "Response from custom pipe"
For more details, check our Plugin Development Guide.