From Linear Steps to Living Pipelines
The real power of FlowManner emerges when you chain agents together. One agent researches, another writes, a third reviews — and the whole pipeline runs in a single click.
What You'll Learn
- Chain agents so output from one feeds directly into the next
- Add conditional branching to handle different outcomes
- Run independent steps in parallel for faster execution
- Monitor workflow execution in real time and debug failures
Multi-step workflows let you decompose complex tasks into discrete stages, each handled by a specialized agent. Data flows forward automatically, and you can branch, merge, or run steps in parallel.
Prerequisites
- At least two agents configured (see Creating Your First Agent)
- Familiarity with the Agents page and Chat interface
- A real multi-step task you want to automate
Sequential vs. Parallel
Sequential steps run one after another — each receives the output of the previous step. Parallel steps run simultaneously and their results are merged downstream. Use sequential for dependent tasks (research → write → edit) and parallel for independent ones (analyze sales data AND scrape competitor prices).
Branching and Conditions
Add conditional branches to route execution based on output. For example, if a content review agent flags quality issues, loop back to the writer. If it passes, send to the formatter. Conditions use simple expressions evaluated against the previous step's output.
Data Flow Between Steps
Each step receives the full output of its predecessor as input context. For long outputs, use a transform step to extract only the relevant data before passing it downstream. This keeps context windows manageable and improves later-step quality.
Error Handling in Pipelines
When a step fails, FlowManner can retry with exponential backoff, skip to the next step, or halt the entire pipeline. Configure error handling per-step based on criticality — a formatting step can be skipped, but a data gathering step should halt the pipeline.
Name steps descriptively
Use names like "Summarize competitor data" not "Step 2". Descriptive names make the execution log readable and debugging much easier.
Watch for context overflow
Long outputs from early steps can bloat the context for later steps. Add a summarization step between stages if you notice quality degrading in later agents.
Test with small data first
When building a new pipeline, test with 2-3 items instead of 100. You'll catch logic errors faster and spend less on API calls during development.
Step-by-Step
Create a New Mission
Go to Missions → New Mission. Give it a name that describes the end-to-end goal, like 'Weekly Client Report Pipeline'.
Add Your First Agent Node
Drag an agent node onto the canvas. Select the agent that handles the first stage of your workflow — typically a data gatherer or researcher.
Connect a Second Agent
Drag another agent node and connect it to the first by drawing an edge between them. The second agent will receive the first agent's output as input context.
Add a Conditional Branch
Insert a condition node between steps. Define a simple rule: if the output contains 'REVISION_NEEDED', route back to the writer; otherwise, proceed to the final step.
Configure Parallel Execution
For independent tasks, add multiple agent nodes without connecting them sequentially. Use a merge node downstream to combine their outputs into a single context for the next step.
Run and Observe
Click Run Mission and watch each step execute in real time. The live stream shows agent reasoning, tool calls, and outputs as they happen.
Related guides
Ready to start building?
Practical tutorials for getting the most out of FlowManner — from your first workflow to advanced multi-agent orchestration.
Start building — it's free