Skip to main content
An Agent is the reasoning core of a Flow — the block that decides what to do. Everything else (Tools, Actions, Triggers, DataSources) exists to give an Agent its inputs and outputs.

​
Agent

An Agent needs three things:
  1. A role — plain-language instructions describing what it should do.
  2. A model — GPT-, Claude-, Gemini-class, or one of Nora’s own. Change any time without rewriting the Flow.
  3. Its Blocks — the Tools it can call, the DataSources it can search, the Actions that run after it.

​
Tool

A Tool is one thing the Agent can call. Examples:
  • An HTTP request to your backend (GET /accounts/{id})
  • A code snippet that computes something
  • A prebuilt integration (e.g., Google Sheets read)
You describe the tool in plain language — Nora generates the calling schema so the Agent knows when to use it.

​
Action

An Action runs after the Agent produces an answer. Think “side effects”.
  • Send an email
  • Post to Slack
  • Write to a Google Sheet
  • Notify a webhook
Actions are optional. A pure Q&A Flow has none.

​
Trigger

A Trigger is what starts the Flow. One Flow, one Trigger (for now). Options include:
  • Chat — a user sends a message from your app or the Nora chat panel
  • Webhook — an external system POSTs a payload
  • Schedule — cron-style, e.g. every hour

​
DataSource

A DataSource is a searchable body of knowledge attached to the Agent. Documents, spreadsheets, or a database. The Agent decides when to search and what to look for. You just point at the source.

​
How they connect

Blocks are wired on the canvas. Typical Flow:
You don’t hand-write the wiring — dragging a connection between two block ports is enough.