NumaSim Docs

Appearance

Chat / AI assistant ​

NumaSim has a built-in chat panel that can read your circuit, propose edits, and explain what's going on. It's powered by a language model and uses the same tool surface the app does — every edit it suggests goes through the same validation as your own clicks.

Opening the chat ​

Click the chat icon in the toolbar (or use View > Chat). A panel opens on the right with a message box and a transcript.

What you can ask ​

Anything you'd reasonably type to a knowledgeable lab partner:

  • "Add a 1 µF capacitor across R2." — the assistant proposes the edit; you click Apply to commit it.
  • "What does this circuit do?" — it summarizes the schematic.
  • "Why is v_mid flat?" — it inspects the latest run and explains.
  • "Convert this to use a Subsystem with N=10 around the switching block." — it refactors and explains the change.

Attaching images ​

You can attach one or more images to a message — a screenshot of a schematic, a photo of a circuit you sketched on paper, or a figure from a datasheet — and ask the assistant to recreate it on the canvas:

  • "Recreate this circuit." (with a screenshot attached)
  • "Build the power stage from this drawing and add a scope on the output."

Three ways to attach an image:

  • Click the paperclip button next to the message box and pick one or more image files.
  • Paste an image straight from the clipboard into the message box (e.g. after a screen capture).
  • Drag and drop image files anywhere onto the chat panel.

Staged images appear as thumbnails above the input; remove one with the × on its corner. A message can carry text, images, or both — an image-only message ("recreate this") is fine.

The assistant reads the image, maps each component to the closest type in the catalog, reconstructs the topology, and builds it with the same tools it uses for a text prompt. Hand-drawn or non-standard symbols are mapped to the nearest catalog part; where a value is illegible or the wiring is ambiguous, the assistant makes a reasonable assumption and calls it out in its summary.

Requires a vision-capable model

Image input only works when the configured model (or Azure deployment) is backed by a vision-capable model. If the deployment does not accept images, the turn fails with an error; switch to a vision-capable model/deployment in the chat settings. Large images are automatically downscaled in the browser before they're sent.

How it works ​

The chat doesn't directly mutate your canvas. Instead it:

  1. Reads a compact description of your circuit (signals, components, wires).
  2. Proposes a diff using the same operation primitives the GUI uses (addComponent, wirePorts, setParameter, ...).
  3. Shows you the proposed diff. You can accept, reject, or refine.

This means every assistant edit is reviewable and reversible — the undo/redo history tracks it just like a manual edit.

Privacy ​

The chat needs to send your circuit description to the model provider to answer. Sensitive circuits should not be shared with the chat. See the app settings for which provider is configured.

Configuring a provider ​

Open the settings dialog and pick a model provider (OpenAI, an OpenAI-compatible endpoint, or a local model). The chat is disabled until a provider is configured.

See also ​