ChatNode
AISend queries to a trained ChatNode agent and receive grounded answers with their source URLs and documents. Agents answer from your own corpus and cite where each answer came from.
What This Integration Enables
ChatNode's connector is deliberately small. Four actions, one trigger. You do not build bots here, you do not upload training files here, and there is nothing in this surface that can delete a knowledge base or reconfigure an agent. You train the agent in ChatNode and you ask it questions from FlowRunner. That narrowness is the point and it is worth saying plainly rather than dressing up: the connector's entire job is to be a trustworthy answering step inside a larger flow.
What makes it worth reaching for over a raw model call is the shape of the response. Send Message returns the answer alongside the source URLs and document links the agent used to produce it. An answer with its provenance attached is a fundamentally different object from a paragraph of confident text. It can be checked in seconds instead of re-researched, it can be refused when the sources are thin, and it can be audited months later. Pair that with Get Chat History and List Conversations, which enumerate what visitors asked and which conversations produced a usable contact, and you have a component that answers from your approved corpus and shows its work every time. Answering is where an agent should be autonomous. Deciding that the answer is fit to send to a customer is where a person belongs, and citations are what make that decision fast enough to actually happen.
Without FlowRunner
With FlowRunner
Use Case Scenarios
First line drafts a human can check without re-reading the wiki
A ticket arrives in Zendesk. The agent lifts the customer's question, calls Send Message against the trained ChatNode agent, and gets back both the answer and the documents it leaned on. It writes the draft into the ticket as an internal note with the sources listed as clickable links. The support rep reads the draft, clicks the one source that matters, sees the paragraph the answer paraphrased, and sends. The reviewing step is a comparison rather than an investigation, which is the only reason a review step survives contact with a busy queue.
A handover audit built from the conversation list itself
List Conversations returns one entry per conversation carrying a flag that says whether the thread was handed over to a live human agent. That single field is a deflection record nobody had to instrument. A nightly flow walks the list for the previous day, splits the conversations into the ones the agent finished and the ones a person had to take, and calls Get Chat History on the handed-over set to read where the conversation turned. The transcripts land in Google Sheets grouped by the question that preceded the handover. Over a few weeks that becomes an ordered account of what the corpus does not cover, assembled from what actually happened rather than from a survey.
Captured contacts routed only when the conversation warrants it
On New Conversation fires with the lead record. With Only With Contact Details enabled it fires solely for conversations that produced a usable name, email or phone, which keeps anonymous browsing sessions out of the pipeline entirely. For each one the agent calls Get Chat History to read what was actually said, summarises the intent, and posts it to the owning rep in Slack. Conversations that read as evaluation intent sync to HubSpot. Conversations the summary cannot classify go to the rep with the transcript attached and a straight question about whether this is a lead at all, because the alternative is a CRM filling with rows nobody trusts.
Human-in-Loop Highlight
Every operation in this connector reads. The irreversible act in a ChatNode workflow therefore never happens in ChatNode at all: it happens the moment an answer becomes a reply a customer receives, in whatever system owns the conversation. Locating the risk precisely is more useful than manufacturing a gate on a read.
Which is exactly why the citations matter. When the agent drafts a reply from a ChatNode answer, it does not post a paragraph and ask "does this look right", because that question has no good answer at speed. It posts the draft with the retrieved sources rendered as links and asks a bounded question: "Draft reply to ticket 8812 on data retention. Retrieved from two sources: the security overview page and the enterprise DPA. Open them and confirm the 90 day figure, or reject." The reviewer checks one number against one document. Answers that come back with no sources at all never reach that queue; the flow routes them straight to a person as an unanswered question, because an uncited answer from a corpus-grounded agent is a signal that the corpus does not cover the topic. The agent does the retrieval and the drafting. A person owns the sentence that reaches the customer, which is the division of labour Orchestration as a Service exists to make routine.
Agent Capabilities
4 actionsAccount
1- Verify API Key Checks that the configured key is valid and active and returns the ChatNode team it belongs to with that team's id, name and slug. Used as a connection health check at the start of an automation, or to confirm which workspace a key reaches before running anything against a bot.
Answering
1- Send Message Sends a query to a trained agent and returns its answer together with any source URLs and document links it used. Pass a Chat Session ID to continue an existing conversation so the agent keeps context; omit it and ChatNode generates a fresh session id for a one off exchange. The returned session id is what Get Chat History accepts later.
Conversations and Leads
2- List Conversations Returns the leads captured by an agent, one entry per conversation with its session id, creation timestamp, the end user's name, email and phone where they were collected, and a flag showing whether the conversation was handed to a live human. It is how session ids are enumerated for Get Chat History and it doubles as a lead export.
- Get Chat History Retrieves the individual messages exchanged between end users and a trained agent, each carrying its chat id, session id, timestamp and a message object typed as either a visitor message or an agent reply. Supply specific session ids, or leave them empty and use a conversation limit to pull the most recent threads.
Triggers
1 triggersEvent Triggers
1- On New Conversation Fires when a ChatNode agent starts a new conversation, emitting the lead record with its session id, timestamp and the end user's name, email and phone where the agent captured them. Enable Only With Contact Details to fire solely for conversations that produced a usable lead, which keeps anonymous browsing sessions out of CRM automations. Follow it with Get Chat History to read what was said. The first cycle records existing conversations without firing, so history is never replayed.
Frequently Asked Questions
What can FlowRunner do with ChatNode?
FlowRunner agents can run Verify API Key, Send Message, and List Conversations in ChatNode, plus 1 more action.
Does connecting ChatNode to FlowRunner require OAuth?
No. ChatNode connects to FlowRunner with an API key, no OAuth flow required.
Can ChatNode trigger a FlowRunner workflow automatically?
Yes. ChatNode supports 1 trigger that can start a FlowRunner workflow automatically.
Start building with ChatNode
$100 in credits. No card required. Connect in minutes.