AI assistant
The AI assistant can set up and change your dashboard for you: add sources and data pieces, write transformers, and review what you already have. You talk to it in the app. It uses a language model from a provider you choose.
Routix Dash does not run a model for you. You supply API keys from your own provider, or you point the app at your own Ollama server. Keys stay in the encrypted configuration on the device.
Each prompt is sent from your device directly to that provider. Routix Dash servers do not receive the chat, the prompts, or the model replies. You are responsible for what you send (including any secrets in the conversation), for that provider’s terms and privacy rules, and for any charges they bill you.
Open the assistant
- On the home screen, tap Ask AI in the top bar.
- Editors can also open the assistant in context, so it already knows which data piece, source, or script you are looking at.
- While Edit layout is on, tap a data piece or component, tap its menu button, and choose AI assistant. The assistant opens for that data piece or component.
Set up a provider
Open the menu → AI settings.
- Choose a Provider: OpenAI, Anthropic, Google, Ollama, or a custom OpenAI-compatible provider (listed by its name, Custom by default).
- Enter the API key from that provider. Ollama does not need a key. For Google, create a Gemini API key in Google AI Studio and paste it here. If Google blocks an older key, create a new one on that page. To keep the key in one place, save it as an ENV variable and enter
{{ env.my_key }}instead of the key. Tap env next to the field to choose a variable or add one, and the app enters it for you. - Pick a Model. For Ollama or a custom endpoint, also set the API URL.
- Use Test connection to greet the model. The reply, or an error, appears at the top of the screen. This check does not use one of your remaining prompts.
Each provider keeps its own API key, Model, and API URL. Switching Provider in AI settings restores that provider’s saved values.
Custom providers
You can keep several OpenAI-compatible providers, for example a local server and a hosted one. In the Provider list, tap + (Add custom provider). Give it a Name so you can tell it apart, then set its API key, API URL, and Model. Delete custom provider… removes the one you are on. At least one custom provider always stays.
Model options
- Reasoning effort — shown when the model supports it: OpenAI and custom providers (for example
gpt-5.6,o3,o4,gpt-oss) and Claude 4.6 and later.- The list shows only the levels that model takes, from Default, None, Minimal, Low, Medium, High, Extra high, and Maximum.
- More effort usually means better answers but slower and more expensive replies.
- GPT-6 models are not offered this setting: on the connection the app uses, they can only use the assistant's actions without reasoning.
- Context window — how much of the conversation the model can take in. It is filled in when you choose a model; unknown models use 256000. When the chat gets close to this limit, it is shortened and the model writes a short summary to carry on from.
- Max tool steps — the most actions (reading your setup, making changes, and so on) the assistant may take for one message. From 1 to 64; the default is 40.
How many prompts you can send
Without a paid plan, the assistant allows up to 150 prompts. AI settings and the message field show how many you have left. Test connection does not count toward that limit. When none are left, the assistant shows Go to plans and the message field is locked until you have Pro.
With Pro, usage is unlimited when you use your own provider keys (or your own Ollama server). You still pay the provider for whatever their service charges.
Without Pro, the assistant also follows the Free limits on data pieces, components, custom screens, and bottom sheets. A change that would go over a limit is refused, and the assistant tells you which limit. See Plans.
What to ask
You can ask it to add a data piece, extract a field, write a transformer, review a section, or fix an error. Short starter prompts appear when the chat is empty.
The assistant can change your configuration. Each time it changes the dashboard, the app saves the change to the history automatically: you see Applying changes, then Applied. One chat is one history entry. It holds a checkpoint of the dashboard from before the first change, a short summary of the work, and the conversation.
The history button in the assistant opens up to 50 entries stored on this device. Tap a row to see the Summary or to restore the checkpoint before the edits. Restoring puts the whole configuration back and returns you to the home screen. You can also restore the conversation to continue it. Swipe a row right to left to delete it.
It sees names and descriptions only — not values. If it needs a password or token, it may ask you to confirm a variable name, or it may open a sheet so you can enter the key, value, and description. The value stays on this device and is never sent to the assistant. See ENV variables.
Before it sends a request, refreshes a data source, or sends data out (including MQTT or WebSocket on a connection you already have), the app asks you to allow that action. You can Allow once, Allow always, or Decline. Always-allowed operations apply to every model and to agents connected through the In-app MCP server. Change them in AI settings → AI permissions.
Snapshots and your configuration
Ask the assistant to save a snapshot, for example "save a snapshot called Before the redesign". It saves your whole configuration as a named snapshot in Settings → Configuration snapshots, where you can restore or delete it. Named snapshots need Pro.
The assistant doesn't show, export, or import your configuration, in whole or in part. To move your setup to another device or keep a copy as a file, use Settings → Configuration export / import. The assistant still explains what a data piece, data source, or transformer does, and changes it for you.
Add images
You can show the assistant a picture, such as a screenshot of an error or a sketch of the layout you want.
- Copy an image, then long-press the message field and choose Paste image.
- On iPad with a keyboard, you can also press Cmd+V in the message field. If text is pasted instead, use Paste image.
Each image appears above the message field as Image 1, Image 2, and so on. Tap ✕ on an image to remove it. A message can have up to 5 images, and you can send images without text.
Images go to your AI provider with that message only; later messages don't send them again. The history keeps only how many images a message had, not the images. If the model can't read images, the assistant tells you; pick a model that can.
Let the assistant check its work
Turn on Render previews in AI settings → AI permissions to let the assistant look at what it builds. It draws a data piece, component, Freeform section, custom screen, or bottom sheet as a picture with your live data and checks how a change looks.
The picture is drawn in the background: your screen doesn't change, and a custom screen or bottom sheet doesn't run its On open or On close actions. The pictures go to your AI provider. It's off by default. While it's off, the assistant asks you in the chat to turn it on when it wants to look.
See Plans for Pro, and Menu and navigation for where AI settings sits in the menu.