Agent flows

Place an agent you published or purchased inside a workflow, where it works like one sealed neuron.

An agent flow is an agent placed inside your workflow. It behaves like one neuron: you fill its inputs and choose where its result goes, but what happens inside stays sealed. Use an agent flow to reuse an agent you published, or one you purchased from the Agent Store, as one step of a bigger workflow.

Add an agent flow

1. Click Add in the canvas top bar, then Agent flow. You can also click Add an agent flow in the Neurons section.
2. In Add an agent flow, find the agent. Type in Search flows... to narrow the list. Each agent shows whether it is Purchased or Published by you, and its version.
3. For an agent you published, choose Add as Copy or Reference. A purchased agent is always added as a copy.
4. Click Add.

The agent flow appears on the canvas in its own frame (AGENT FLOW) as a sealed neuron, with an AGENT tag in the Neurons section. Its Get tab opens so you can fill its inputs.

Copy or Reference

• Copy: the workflow gets its own copy of the agent as it is now. Later changes you publish to the agent don't change this copy.
• Reference: the workflow uses your published agent itself, so it follows the updates you publish.

Use an agent flow in a workflow

An agent flow's settings panel has four tabs: Start, Get, Agent and Send:

• Start: when the agent flow runs, as with any neuron.
• Get: fill the agent's inputs under Value to run with, or connect another neuron to fill them.
• Agent: shows The sealed flow this agent runs and its last result. The neurons inside are sealed: the agent's Builder controls them; you control the inputs and where the result goes.
• Send: choose where the result goes, as with any neuron.

Connect an agent flow to other neurons the same way as any neuron. See Connections.

What a connection without a link fills

If you connect a neuron to an agent flow without linking its output to a particular input, the output goes into the agent's first empty input of the same type (text into a text input, a photo into an image input). Inputs the agent's Builder locked (set in advance and shown as Can see) are skipped, and an input that already holds a value is never overwritten. To choose the input yourself, link it in the connection's settings.

The run stops before the agent starts, with a message, when:

• the agent has no input of that type (the message names the agent);
• every input of that type already holds a value (the message names the agent): clear one and run again;
• the output can't be combined with the agent's other inputs.

Remove an agent flow

In the Neurons section, open the agent flow's menu (Run and options) and choose Remove agent flow. The same menu has Play, Agent details, Collapse on canvas and Rename.

Common problems

An agent shows offline in the list

An agent marked offline in Add an agent flow isn't available right now and can't be added. Pick another agent, or try again later.

An agent can't be added to a workflow

If Add shows a message that the agent can't be added to a workflow, the agent was published under older rules that it no longer meets, for example more than 5 inputs inputs. Nothing is added to your workflow. The agent's Builder needs to publish it again; you can still run it from My agents if you added it there. See My agents.

No published or purchased agents yet

The agent flow list shows only agents you published or purchased. Publish a workflow, or add an agent from the Store, then try again. See Agent Store.

Credits for an agent flow

An agent flow's runs use credits like any other run. If you purchased the agent, each run of the agent flow also adds what its Builder charges per run, if anything: a share of that run's cost and any fixed fee per run. Both are paid to the Builder. Agents you published yourself add neither. See Pricing & earnings.

Copy and Reference aren't offered

A purchased agent can only be added as a copy. Copy and Reference are offered only for agents you published yourself.