🌐 Workflows & Swarm

Swarm Canvas & Multi-Agent Orchestration

Build complex multi-agent workflows with the drag-and-drop Swarm Canvas. Route logic, conditionals, parallel execution, and simulation mode.

Updated May 13, 2026

6 min read

The Swarm Canvas is FuseIQ's visual workflow builder. Drag, drop, and connect AI agents, set routing conditions, and orchestrate complex multi-agent workflows without writing code.

Canvas Overview

The canvas starts with a clean grid where you add nodes representing agents, logic gates, and output destinations. Key components:

Input Node — Entry point for the workflow (manual trigger, webhook, cron, or event)
Agent Node — Any connected agent in your workspace. Configure prompts and parameters.
Condition Node — Route workflow based on agent output content, cost, or status
Output Node — Send results to integrations (Slack, Discord, GitHub, etc.)
Parallel Gate — Execute multiple agents simultaneously and merge results

Building a Workflow

Open the Swarm Canvas from the sidebar

Drop an Input Node onto the canvas. Configure the trigger type (webhook, cron, or manual).

Connect an Agent Node. Select which agent runs, set its system prompt, and define expected output.

Add a Condition Node after the agent. Example: route to "human review" if confidence is below 80%.

Connect an Output Node to post results to Slack, email, or another integration.

Run a Simulation to test the workflow before deploying.

Advanced Patterns

Beyond simple linear workflows, the Swarm Canvas supports:

Parallel Execution — Use a parallel gate to run multiple agents at once. Example: research, write, and review simultaneously.
Conditional Routing — Route based on token cost, execution time, or output content. Example: expensive queries go to a cheaper model.
Loop & Retry — Automatically retry failed agent nodes with configurable backoff.
Sub-Flows — Nest entire workflows as a single node in another canvas.
Human-in-the-Loop Gates — Pause execution at any node until a human approves or edits the agent's output.

Simulation Mode

Before deploying a workflow to production, use Simulation Mode to:

Step through each node with sample input data
Verify agent outputs match expected formats
Check condition routing logic
Estimate execution cost and time before it runs live

The canvas enters a simulation overlay with a coloured tint and debug panels showing each node's input, output, and cost. Run unlimited simulations — they don't count toward your monthly execution quota.

💡

Use simulation to A/B test different agent configurations before settling on a pipeline.

Output Integrations

Workflow outputs can be sent to any of 18+ platforms, including Slack, Discord, GitHub, Jira, Google Sheets, email, and custom webhooks. See the full list in the Output Integrations article.

Was this helpful?