agenthub · input.chat
Chat Input
Where a conversation enters the graph. It takes whatever the channel handed the run and turns it into the three shapes the rest of the canvas expects.
Why the node earns its place
Every chat or voice agent starts here, and it exists because the thing that starts a run is not shaped like a conversation. A WhatsApp webhook, a test run from the builder and an SDK call all deliver the message differently — flattened under a node id, nested under inputs, or as a bare string.
This node absorbs that. Downstream nodes never learn there was more than one convention, which is why the same graph runs unchanged whether it was triggered by a person messaging on WhatsApp or by you pressing Test Run.
How it works at run time
- 1Look for the message under four conventions in order: nested by node id, nested by node name, flat by node id, flat by node name — then fall back to a plain
textkey, or a bare string. - 2Normalise what it found. A string becomes a user message; a dict carrying
textis used as the message object directly. - 3Emit all three shapes at once, so the next node picks whichever it needs.
What comes out
text "is my renewal due?" // simple pipelines
message {role: "user", content: …} // structured workflows
messages [{role: "user", …}] // straight into an LLMWhat you wire
Inputs
Nothing to wire in.
Outputs
- messageobject
- textstring
- messagesarray
Configuration
| Key | Type | Default & options | What it does |
|---|---|---|---|
| label | string | default "User Message" | — |
What usually goes wrong
Watch for this
When nothing resolves, text is an empty string and messages is an empty array — not an error. An empty array reaching a model is rejected by the provider, so if a Test Run fails with an invalid-messages complaint, the fault is almost always here rather than at the LLM.
Behaviour & provenance