Build your first flow

You will build a flow that listens for a webhook, sends the payload to an LLM, and returns the model’s reply as the HTTP response. By the end, you will have a real, addressable URL you can curl from your terminal.

Before you start

You need a Glyph account and one secret configured: an OpenRouter key. If you do not have access yet, ask your admin.

1 · Create the document

Open glyph.naburis.cloud/app/ (the left sidebar opens on Home). Click New Board in the header, then in the creation modal click Browse Templates and pick Map a Workflow — a flow-mode board with a starter graph. Rename it to Echo Bot in the editor’s title bar; autosave handles persistence (there is no save button).

2 · Replace the trigger

The template seeds a Manual Trigger. Select it and press Delete. Then open the operator palette: click Insert on the left rail, or press Tab. Filter to the Trigger family and click Webhook Trigger to add it to the canvas. In the inspector (right side, Setup tab), leave the method as POST. Glyph mints the trigger URL as soon as the node lands.

The URL appears in the trigger’s inspector under Endpoint. Copy it. It will look like https://glyph.naburis.cloud/api/triggers/<id>.

3 · Add an LLM call

Open the palette again (Tab), filter to the Model family, and click Respond to drop it onto the canvas. In its inspector, set:

• Model: deepseek/deepseek-r1
• System prompt: You are a concise assistant. Reply in one sentence.

4 · Wire them up

Drag from the trigger’s payload output port to the Respond operator’s prompt input port. The edge should turn green — that means the port types match.

5 · Add the response

Open the palette (Tab), filter to the Output family, and click Return to add it. Wire the Respond operator’s output port to its input.

6 · Trigger it

In a terminal:

curl -X POST <your-trigger-url> \
  -H 'content-type: application/json' \
  -d '{"prompt": "Why is the sky blue?"}'

You should see a one-sentence answer in the response body. In the canvas, the run badge on the trigger turns green and the run timeline appears in the Run console at the bottom.

✓

What just happened

You composed a flow from three operators, the backend compiled it into a DAG, the worker received the run when your curl hit the trigger URL, and the executor ran the three nodes in dependency order — exactly once, idempotently, with the run state persisted at each checkpoint.

Where to go next

• Open the latest run in the Run console and click through its checkpoints — each one is a row in the database.
• Pin a sticky note next to the LLM node explaining your prompt choice. (You’re now using the document in hybrid mode.)
• Read Why operators — the conceptual reason your three nodes wired up cleanly.

For the full operator surface — config tables and port lists for every family — see Operators by family.

If you want to know how to build a new operator from scratch, that’s an admin-tier tutorial: Build your first operator.