Agentic workflows

Agentic workflows are how amAIra goes beyond a general chat. You ask in ordinary language, or you run a short slash command. The assistant uses the scene you are actually looking at.

Antinode Antinode
amAIra amAIra

What you need

Requirement Why it matters
Sign in on the home page Chat uses your Antinode account. See Antinode
Infrastructure on Connected A skill that reads or runs on the simulation calls the server
Something loaded Answers refer to the current scene and time step

If the amAIra tab is missing, sign in on the home page and open the viewer from there.

The pieces

Piece What it is for
Antinode Sign-in, the account, and the platform behind amAIra
amAIra The chat tab on the right
Knowledge Notes the assistant can read: how a Scenaven session works, your own cheat sheet, or context for your discipline
Skills Actions you can call, often with a slash command, that read live data or run an approved step on the simulation
Scripts Short Python a skill can run on the server, with the same limits as the Python session. You choose the skill in chat. The skill runs its script

Shared skills — describe the session, a field report, compare frames, and similar — are already available after you sign in. A skill you add yourself works the same way.

Shared and personal

Kind Who it is for
Shared Workflows Scenaven provides. Everyone who signs in can use them
Personal Workflows you add. Only you see and edit them

In the amAIra panel, open Workflows (next to Chat) to add your own Knowledge (a markdown note) or Skills. If a skill should run Python, upload that script in Workflows as well and attach it to the skill. Your items are used together with the shared ones. Autocomplete shows which is which.

You can chat without uploading anything. Shared guidance may already be there after you sign in.

What the assistant uses

A shared skill and a skill you added run the same way. You ask, or you type / and pick a skill. The skill runs on the server against the simulation you have open. The answer shows in the chat, sometimes with a hint in the viewer.

You can leave field names and frame numbers out. The assistant uses the current selection unless you ask about something else. If the skill needs a detail, it asks for that — a field, a frame, or which object.

The assistant updates when you:

  • Load an example or open a file
  • Select a different row in Scene
  • Move to another time step
  • Apply a filter or start a new scene

Ways to use it

Ask in plain language

  • “What fields are on this mesh?”
  • “Summarize this transient dataset.”
  • “What clip would isolate the high-stress region?”

The answer mixes Knowledge with the live session when a matching Skill exists.

Slash commands and @ references

  • Type / to pick a Skill, for example to describe the session or compare two frames.
  • Type @ to pull a Knowledge note into the message by name.

A slash command runs on the server against the simulation you have open, and the result comes back in the chat.

Chat and the Python session

Workflows answer and suggest. The Python session is where you edit and run a snippet yourself. A skill that runs Python follows the same limits: no import, and the names session, np, and vtk. A common pattern is to ask in chat, then try the snippet the assistant offered.

Troubleshooting

What you see What to check
No amAIra tab Sign in on the home page and open the viewer from there
Chat works, but / lists nothing Open Workflows and see what is published. Shared skills may not be available yet
A personal upload fails You need to be signed in. Knowledge is a markdown note. A skill that runs Python needs its script uploaded in Workflows and attached to that skill
Answers feel generic Shared notes may be broad. Add your own in Workflows
A skill fails or times out Infrastructure should say Connected, and a dataset should be loaded
A skill says it cannot run its script In Workflows, confirm the script is uploaded and the skill is attached to it
The answer is about the wrong object or frame Select the row or time step you mean, then ask again

Next steps