Introduction
What Octaviaflow is, how the pieces fit together, and how it differs from a plain integration tool.
Octaviaflow is one platform for two jobs that are usually split across two tools: moving data between systems, and automating the work that happens around that data.
A traditional integration tool syncs records on a schedule. A workflow tool reacts to events and calls APIs. Octaviaflow does both in the same canvas, so a nightly sync and the "if the total is over 10,000, ask someone" step live in one place rather than two products that cannot see each other.
How the pieces fit#
Five things make up everything you will build.
Piece | What it is | Where it lives |
|---|---|---|
Connection | A saved, authenticated link to one of your systems | Apps page |
Flow | A graph of steps on a canvas — your automation | Inside a project |
Trigger | The one step that decides when a flow runs | First node of a flow |
Action | Every other step: read, write, branch, loop, call, extract | The Action library |
Agent | A reusable AI worker you drop into a flow as a step | AI → Agents |
The relationship that matters most: a connection holds authentication, a flow holds behaviour. You authorise Salesforce once. After that, a sync flow, an alerting flow and a reporting flow all use that same connection while doing completely different things. Rotating the credential fixes all three at once.
What a flow actually does#
Every run starts with the trigger producing a trigger envelope — the schedule that fired, the webhook body that arrived, the record that changed. Each step then reads the output of the steps before it and writes its own output under its own step ID.
That is the whole model. There is no hidden global state: if a step can see a value, it is because an earlier step produced it. Data References covers exactly how to address those values.
Where the AI part comes in#
AI is a step type, not a separate product. An AI Agent is configured once — instructions, model, optional tools, optional knowledge base — and then referenced from as many flows as you like. Because it is a normal step, its output maps into a Transform, gets tested by a Condition, or is written to a destination like any other data.
Agents can also be given structured output, so what comes back is typed fields rather than prose you have to parse. See AI Agents.
What it is not#
Being clear about the edges saves time:
- Not a reverse-ETL warehouse product. Flows move records between operational systems. There is no modelling layer.
- Not a general-purpose code runtime. Expressions and Custom Functions cover computation; anything heavier belongs behind an HTTP endpoint you call.
- Not a BI tool. Query can aggregate, and you can write results anywhere — but the charts live in whatever you already use.
Next#
Quickstart builds a running flow in a few minutes. Core Concepts defines the vocabulary the rest of these docs assume you know.