Agent.ai
AIConnect AI agents to the Agent.ai Actions API for model calls, web research, and company or person enrichment. Agents run published Agent.ai agents as a step inside a larger FlowRunner workflow.
What This Integration Enables
Two thirds of this connector is a competent toolbelt. Generate Content reaches models from several providers through one credential. Get Web Page Content and Get Search Results ground an answer in live pages rather than in training data. The enrichment actions fill in a lead record from whatever identifier arrived first, whether that is an email, a company domain, a ticker symbol or a LinkedIn URL. That is useful and it is not unusual. FlowRunner runs a BYOK model for AI providers precisely so teams can hold their own relationships with model vendors, and Agent.ai is a different shape of the same instinct: one key sitting in front of several providers so a model comparison does not require a procurement cycle.
The last third is the part worth thinking about. Describe Agent and Invoke Agent let a flow call an agent that somebody else built and published, and Invoke Action calls any action in the live Agent.ai catalogue by name with a free-form parameter object, which covers far more than the actions this connector wraps explicitly. That is genuine leverage: the fastest way to get an agent's job done is often to run one that already exists, which is the same reasoning behind FlowRunner's Agent Directory. It is also a dependency, and dependencies in a production path deserve the same treatment as any other. This is where human-in-the-loop belongs on this connector: not on the model call, on the decision to trust a piece of logic you did not write.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Enriching a lead before anyone spends time on it
A form submission arrives with a work email and nothing else. The agent calls Enrich Person with the email, then Get Domain Info on the company domain to check registration age, then Get Company Financial Profile when the company turns out to be publicly traded. Within a few seconds the record in HubSpot carries a role, a company size signal and a domain age, and the routing rule has something real to work with. Leads that come back thin are held for a person rather than assigned to a rep who will waste an hour.
A daily digest built from current sources
A scheduled flow calls Get Google News with a query and a one-day window, then Get Web Page Content on the two or three results that matter, then Generate Content to write a short digest that quotes the sources it read. The digest posts to Slack each morning. When the news window returns nothing worth reporting, the flow posts nothing, which is the behaviour a team actually wants from a digest.
Running a published agent as one step in a larger flow
A competitive research agent already exists on Agent.ai and does the job well. Rather than reimplementing it, the flow calls Describe Agent to learn exactly which input variables it expects and what it returns, maps the flow's own fields onto those variables, and calls Invoke Agent for a finished result in a single synchronous response. The output feeds a scoring step and lands in Notion. The logic stays where its author maintains it, and the flow stays short.
Human-in-Loop Highlight
Invoke Agent and Invoke Action are the two operations where an agent should stop and ask, and the reason is unusual enough to state plainly. Everywhere else on this page, the risk is a wrong value written to a record. Here, the risk is that the flow's behaviour changes without the flow changing. Invoke Action calls any action in the live Agent.ai catalogue by name with free-form parameters, so what a step can do is defined by a catalogue you do not control. Invoke Agent runs logic authored and maintained by someone else. Neither is a reason to avoid them, and both are a reason to review them once, on purpose. So before a published agent goes into a production path, the flow pauses and posts what Describe Agent returned: "Proposed step: agent research-brief-v3, published by an external author, 41,000 executions, review score 4.6, expects a company domain and returns a summary plus a source list. This step will run against live CRM records. Approve this agent for this flow, or pin to the wrapped actions instead?" A person accepts the dependency by name, the approval is recorded, and six months later there is an answer to the question of who decided to call it.
Agent Capabilities
16 actionsAI Generation
3- Generate Content Sends instructions to a large language model through Agent.ai and returns the generated text in one response. A single key reaches models from several providers, and a model identifier that is not in the list can be passed through unchanged. Used for summarising, drafting and classification inside a flow.
- Generate Image Creates an image from a text description, with optional control over the model, visual style and aspect ratio. Used to produce illustrations, thumbnails or social assets as part of a content workflow.
- Generate Audio Converts a block of text into spoken audio and returns the generated output. Used to turn summaries, articles or notifications into voice content.
Web Data
5- Get Web Page Content Extracts the readable text of a web page along with metadata such as title, description and favicon. Single Page reads only the given URL, while Crawl Site follows links across the same site up to a bounded number of pages. Used to ground a model on a live page instead of on training data.
- Get Web Page Screenshot Captures a visual screenshot of a web page and returns it as an image, with an optional cache lifetime so repeated captures of the same page reuse a recent render. Used for visual evidence in review and monitoring steps.
- Get Search Results Runs a web search and returns the organic result list with titles, links and snippets. Used as the first step in any workflow that should answer from current web content rather than model memory.
- Get Google News Fetches recent news articles matching a query, each with title, source, publication date, link and snippet, restricted to the last day, week or month. Used for press monitoring and daily digests where recency is the point.
- Get YouTube Transcript Retrieves the spoken transcript of a YouTube video so it can be summarised, searched or repurposed. The video must have captions available. Used to bring recorded content into a research workflow.
Enrichment
4- Enrich Person Looks up a person and returns an enriched profile from whichever identifiers are supplied, accepting any combination of email, name, company, domain or LinkedIn URL. Used to fill in a lead record before outreach, where more identifiers generally produce a better match.
- Get Company Financial Profile Returns financial and company profile data for one or more publicly traded companies, such as market capitalisation and last known stock price, from a comma-separated list of ticker symbols. Used to size an account before it is routed.
- Get Domain Info Returns the registrar record for a domain, including registration and expiry dates, name servers and published registrant details. Used to vet a lead's company or check domain age as a quality signal.
- Get LinkedIn Profile Retrieves the public details of a LinkedIn profile, including headline, current and past positions, education and certifications, from the full profile URL. Used to confirm a contact's role before an outreach step commits to a message.
Published Agents and Actions
4- Describe Agent Returns the full metadata of a published Agent.ai agent by ID or slug, including its name, description, type, tags, execution count, review score and the input and output variables it works with. Used as the review step before any agent is wired into a flow, and to learn exactly what input it expects.
- Invoke Agent Runs a published Agent.ai agent and returns its finished output in a single synchronous response. Used to reuse an existing agent's logic as one step rather than rebuilding it, after Describe Agent has established what it takes and returns.
- List Actions Lists every Agent.ai action available to the API key, each with its name, description and a JSON schema of the parameters it accepts. Used to discover actions this connector does not wrap explicitly, and to learn the exact fields to pass to Invoke Action.
- Invoke Action Calls any Agent.ai action by name with a free-form parameter object, covering the whole live catalogue rather than only the wrapped actions. Used for capabilities with no dedicated action here, with the parameters taken from what List Actions reports rather than assumed.
Frequently Asked Questions
What can FlowRunner do with Agent.ai?
FlowRunner agents can run Generate Content, Generate Image, and Generate Audio in Agent.ai, plus 13 more actions.
Does connecting Agent.ai to FlowRunner require OAuth?
No. Agent.ai connects to FlowRunner with an API key, no OAuth flow required.
Can Agent.ai trigger a FlowRunner workflow automatically?
Agent.ai doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Agent.ai
$100 in credits. No card required. Connect in minutes.