Pipeline

A Pipeline is a sequence of stages that transform data. Pipelines are defined in YAML and executed by the Orchestrator.

Structure

name: "Voice Assistant"
registry: "http://localhost:8080"

input:
  kind: "AudioRaw"

stages:
  - whisper-tiny@1.0
  - target: integration
    provider: openai
    model: gpt-4o-mini
  - kokoro-82m@0.1

How Pipelines Work

Stage Formats

Simple Format

Reference a model by ID and version:

stages:
  - whisper-tiny@1.0
  - kokoro-82m@0.1

Object Format

For explicit configuration:

stages:
  - name: whisper-tiny@1.0
    target: device

Integration Stages

For cloud LLM execution:

stages:
  - target: integration
    provider: openai
    model: gpt-4o-mini
    options:
      system_prompt: "You are a helpful assistant."
      max_tokens: 150
      temperature: 0.7

Execution Targets

Target	Description	Source
`device`	On-device inference	.xyb bundle from registry
`server`	Cloud inference	Xybrid cloud (future)
`integration`	Third-party API	OpenAI, Anthropic, etc.
`auto`	Framework decides	Resolved at runtime

Input Types

Declare expected input type for validation:

input:
  kind: "AudioRaw"   # For ASR pipelines

input:
  kind: "Text"       # For TTS or text pipelines

Registry Configuration

Simple URL

registry: "http://localhost:8080"

File Path (Local)

registry: "file:///Users/me/.xybrid/registry"

Data Flow

Each stage transforms an Envelope:

Stage Type	Input	Output
ASR (Whisper)	`AudioRaw`	`Text`
LLM (GPT-4o)	`Text`	`Text`
TTS (Kokoro)	`Text`	`AudioRaw`

The Pipeline validates that outputs match next stage's expected input.

Lifecycle

Load - Parse YAML, resolve models
Validate - Check type compatibility
Execute - Run stages via Orchestrator
Unload - Release resources

Pipeline DSL - Complete YAML reference
Orchestrator - Executes pipelines
Registry - Model bundle resolution

Pipeline

On this page