Agent marketplace

What an agent is

A marketplace agent is a ready-made workflow for one business job: a trigger, AI steps with expert-written prompts, and the apps it uses (for example Slack or HubSpot). Setting it up creates that workflow in your workspace; from then on it's a normal workflow you can open, edit, test and monitor.

Browse the marketplace

Open Agents from the sidebar. Each card shows the agent's name, what it does, and either its one-time price or "Included in your plan" / "Purchased". Press Details for the full picture: what it does, what it needs (plan, AI key, apps), what it doesn't do, its steps and roughly how long setup takes. Filter by category, search by name, or tick "Available now only".

Available, coming soon

• Available: the agent's workflow is built and tested, and every app it needs is live. You can buy it and set it up.
• Coming soon — waiting for <app> approval: the agent is built, but an app it needs is still waiting for the vendor's approval. It becomes available automatically once that app is approved.
• Coming soon: the agent's ready-made workflow is still being built. It can't be bought yet.

Categories

The marketplace has 219 pre-built agents in 30 categories; 204 of them are available now (the Agents page in the app shows the current status). Every agent has a setup guide — see “Setup guides for every agent” below.

Buying agents

• Agents need a paid plan: they run live, and the Free plan only runs Dry Runs. On Free the button says "Upgrade to use".
• Paid agents are a one-time purchase per workspace: press Buy on the card and pay through Stripe checkout. There's no subscription.
• Some agents are included with your plan: Pro and Business include 2, Elite includes 5, and Enterprise includes all of them. Included agents show "Included in your plan" and are yours without buying.
• Agents marked "Coming soon" can't be bought until they're available.
• Agents run on your own AI provider key, like every other AI step; model usage is billed by your provider.

Setting up an agent

• In Agents → My agents, press "Set up this agent". It creates the agent's workflow in your workspace, named after the agent, with its AI steps set to a model from the AI provider you connected. It counts towards your plan's workflow limit.
• The canvas opens with a setup checklist above it. Each item links to the exact place: Settings → AI Providers, the app to connect in Settings → Integrations, the step and field to fill in (e.g. the Slack channel), Dry Run, and the trigger's Activation panel.
• Items tick themselves off as you go: settings as soon as you fill them in, the rest when you come back to the canvas or after a Dry Run.
• Dry Run uses the agent's sample data (the trigger's "Sample data for Dry Run" box) and sends nothing.
• When every item is done and the trigger is on, the checklist says "Live — this agent is running". Follow its runs in History and in Analytics → Agents (runs, success rate, last run, AI cost).
• Already set up? My agents shows "Open workflow" and the setup progress instead.

Every agent has a step-by-step setup guide at /docs/agents/<agent id> (linked from the agent's Details in the app and listed below by category): what it does and doesn't do, the apps to connect, every setting to fill in, how its workflow runs, how to test it with Dry Run and how to switch it on.

How agents start

• From a webhook: your form, chat widget or app sends each new item to the trigger's webhook URL.
• On a schedule: for example every weekday morning, in the trigger's timezone.
• From an app event: something new in a connected app (a sheet row, a booking, a payment…). The app must be connected, and the trigger's Activation panel needs that account before "Activate app trigger" works. Steps read the event as {{trigger.<field>}}.
• Some agents also wait between steps (a Wait step, for follow-ups), repeat steps for each item of a list (For each item), or remember earlier conversations (Memory). See the Node reference for how these behave.

Model selection

Building your own agent step

• There's no separate agent builder: add an Agent node on the canvas and set its prompts, model, temperature and tools.
• The built-in tool today is an HTTP request tool (call a URL with arguments the model fills in).
• Need memory, lookups, email or another workflow? Chain the Agent with Memory, Retriever, Output or Sub-workflow nodes.

Common errors & fixes

Output is too generic

Lower the temperature (0 = most consistent, 1 = most creative), add examples to the system prompt, or switch to a stronger model.

Agent makes up facts

Give it the facts: add a Retriever or HTTP step before it and include the results in the prompt, then tell it to only use the information provided. Add a Validator after it to check the output.

Output blocked unexpectedly

A Guardrail step may have matched a banned term or field. Open the Trace tab to see which step stopped the run.

Setup guides for every agent

219 agents in 30 categories. Each guide is generated from the agent's own workflow: what it does and doesn't do, what to connect, every setting to fill in, how to test it with Dry Run and how to switch it on.

Available Coming soon (waiting for an app's approval) Coming soon