Orchestrate multiple AI agents in parallel on Invoked's visual canvas — chain MCP tool calls across agents for complex multi-step workflows.

What is the canvas?

The Multi-Agent Canvas is a visual workspace where you can create multiple agent nodes, connect them, and run them in parallel or in sequence. It's designed for workflows that are too complex for a single agent — where different subtasks benefit from specialization.

Adding agents to the canvas

Open the Canvas panel (⌘5)
Click Add agent or press A
Configure the agent's role, model, tools, and prompt
Repeat for as many agents as you need

Connecting agents

Draw connections between agents to define information flow:

An arrow from A → B means B receives A's output as context
Agents without incoming connections start immediately when you run the canvas
Agents with incoming connections wait for their dependencies to complete

Running the canvas

Click Run all to execute the canvas. Agents run in dependency order — independent agents run in parallel, dependent agents wait.

You can also right-click any agent and choose Run this agent to execute it individually.

Canvas layouts

Use the Layout menu to auto-arrange your agents:

Horizontal — left to right, dependencies to the right
Vertical — top to bottom
Auto — Invoked infers the best layout based on your connections

Example: PR review pipeline

[Fetch PR]          [Fetch linked issues]
     ↓                       ↓
     └──────────┬────────────┘
                ↓
         [Summarize & score]
                ↓
        [Write review comment]

Four agents, three of which can run in parallel (the first two), with the final two sequential.

Limitations

Maximum 20 agents per canvas
Circular dependencies are not allowed
Canvas state is saved automatically but runs are not persistent across restarts

Multi-Agent Canvas