Concepts & terminology

The basics

Workflow

One automation: something that starts it, steps that work it out, and somewhere the result goes. Drawn as connected cards on the canvas.

Canvas

The drawing surface where a workflow's steps live. Click a card to open its settings in the panel on the right; drag between the dots to connect steps.

Step (or node)

One card on the canvas: one job. Steps have plain names — “Decide what matters”, “Send it to Discord” — and every one AWO writes arrives already named and configured.

Simple and Advanced

Two readings of the same workflow. Simple speaks plainly and hides machine words. Advanced shows the file as it is — ids, raw settings, exact values — and its promise is that the panel matches the file one-for-one. Flip freely; nothing changes but the words.

Inputs — what starts a workflow

Input (trigger)

The card a workflow starts from. A workflow can have several; each event runs only the branch its own input reaches.

When a page changes (page watch)

Checks a web page on your schedule and runs only when it changed. Steps read what changed — the added lines — not the whole page.

When something calls in (webhook)

The workflow gets its own web address; anything that can send a message posts events to it. The event's content is “the payload”.

Collect a group (batch delivery)

Instead of running once per event, hold events until a ceiling — so many events, or so much time — then run once over the whole group. One digest, not fifty pings.

On a schedule

Runs at a fixed interval, changed or not.

Only when you run it (manual)

Runs when you press the button, never on its own.

Stand-in

A placeholder in a template for something only you can know — which page, which channel. The bar above the canvas names every stand-in still unanswered; press one and the cursor lands on it. A workflow with stand-ins can test-run, but AWO refuses to set it watching until they're real.

Steps — working it out

Ask the AI

Hands what arrived to your model with your question and keeps the answer. The step most workflows are built around.

Typed pieces (the output contract)

A switch on the AI step. Off, the AI answers in one piece of text. On, it returns display-ready pieces — a number, a table, a status label, a chart's worth of data, a paragraph — that dashboards draw and channels format. AWO tells the model the exact shape and checks the reply, so formatting is never your problem. If a run produces fewer pieces than usual, the missing ones simply leave their widgets unchanged.

Only continue if… (condition)

A gate: checks one value, and the steps after it run only when the check passes. A closed gate is a quiet success, not an error.

For each item… (loop)

Repeats the steps after it once per item in a list.

Pull out data (extract)

Grabs a piece of what arrived — no AI involved.

Send output

The last step: where the result goes. A dashboard, a message to Discord/Slack/Telegram or email, a web request to another app — or a dashboard and a channel from the same run.

Dashboards

Board (dashboard)

A page of widgets your workflows fill. Create one yourself, or let a workflow create it — AWO picks widgets that fit what's published.

Widget

One tile on a board: a number, a chart over time, a paragraph, a status badge, a table, a feed of everything found. Drag to move, resize, press ⚙ to change what it shows. A widget that a running workflow fills says so by name while it waits for the first result.

Publishing and routing

Two ways results land on boards: a workflow publishes named fields (price, summary, status) that widgets read, or it routes typed pieces straight into specific widgets. Recipes and drafts wire this for you.

Running and watching

Test run

A real run — the model is called, messages really send — where only the starting event is invented for you. How you see a workflow work before trusting it.

Run

One execution, recorded step by step: what each step received, decided and produced, under its own name. The Runs screen keeps them all.

Watcher (watching)

A workflow set running on its own — checking its page, listening at its address, keeping its schedule — even with the browser closed. The Watching screen shows each one's truth: last ran, next check, and a plain reason whenever one can't run.

Check for problems

Inspects the workflow before anything runs: missing values, steps reading things that can't exist, deliveries with nowhere to go. Problems state their fix — many carry a button that applies it.

Models and Setup

Model

The AI that answers your workflows' questions. Local (Ollama — found automatically if it's running) or an account with OpenAI or Anthropic. Picked once in Setup; any workflow can override it.

Connector

The saved details AWO uses to reach a model — the machine word for “which AI, where, with what key”. You'll only meet it in Advanced.

Setup

The first-run wizard and its settings: your model, and optionally the channel your alerts go to. Changeable any time from the menu.

Ready-made and AI-made

Recipe

A fill-in-the-blank workflow: answer two or three plain questions and AWO writes the rest, dashboard included.

Template / example

A complete workflow you start from and adjust. The gallery ships dozens; you can save your own.

Describe it (the drafter)

Type what you want in a sentence; AWO drafts the workflow onto the canvas for review — steps named, dashboard created, nothing run until you say so.

Explain

The button on the workflow summary that has your model walk through what the workflow does, in plain words, grounded in what each block actually does.

Project Autopilot

Project

A goal handed to AWO to plan and build: a repository, a goal in your words, and optionally which files are in scope.

Phase

One chunk of the plan — a title, what it's for, and “done when” criteria. Phases wait on the phases before them; the board shows each one's state.

Cycle

One pass of work inside a phase — research, plan, implement, review, test — the loop AWO runs until the phase's criteria hold.

← Back to the guide