Flezi AgentBox

Pipelines Guide

Create multi-agent pipelines with the visual DAG builder, validation, and publishing

Pipelines Guide

Agent Pipelines let you chain multiple agents together for complex, multi-step workflows. This guide covers creating, validating, and publishing pipelines.

What Are Pipelines?

A pipeline connects multiple published agents in sequence or parallel, passing data between them automatically. Each agent in the pipeline receives the previous agent's output as input.

yaml
# Example: Code Review Pipeline
Pipeline: Code Review
  Stage 1: code-analyzer (Sequential)
  Stage 2: security-scanner (Parallel with Stage 3)
  Stage 3: style-checker (Parallel with Stage 2)
  Stage 4: report-generator (Sequential, waits for 2 & 3)

Creating a Pipeline

Visual Pipeline Builder

Navigate to your dashboard and click Create Pipeline:

Name your pipeline

Give it a descriptive name and category.

Add agents

Search for published agents and drag them onto the canvas.

Connect agents

Draw edges between agents to define execution flow.

Configure transformations

Set up data transformation rules between agents if their input/output formats differ.

Validate

Click Validate to check for issues before publishing.

Execution Modes

Sequential Execution

Agents run one after another. Each agent receives the previous agent's complete output before starting.

Parallel Execution

Independent agents run simultaneously. A synchronization point (join node) waits for all parallel branches to complete before continuing.

Parallel branches must not depend on each other's output. If Agent B needs Agent A's output, they must run sequentially.

Data Transformation Rules

When connecting agents with different data formats, configure transformation rules:

  • Field Mapping — Map output fields to input fields
  • Format Conversion — Convert between JSON, text, and structured data
  • Filtering — Select only relevant fields from the previous output
  • Merging — Combine outputs from parallel branches

Validation

Pipeline validation checks for:

  • Cycle Detection — No circular dependencies allowed
  • Schema Compatibility — Output format matches next agent's expected input
  • Reachability — All agents accessible from the start node
  • Agent Availability — All referenced agents are published and active

Publishing Pipelines

Published pipelines appear in the marketplace alongside individual agents:

  • Set pricing (per-execution or free)
  • Add description and documentation
  • Manage versions with semantic versioning

Next Steps