Setup

The very first time you open Trustant you land on the splash screen. Before you can create an application, Trustant needs to know which AI provider powers the assistant — the model that turns your descriptions into working code.

This is the only mandatory setup step. Everything else has a sensible default.

Choosing a provider

The Select your AI provider screen offers two ways to get a model, and the difference between them is where the model runs and who pays for it:

Whichever you pick, the choice is saved in your workspace configuration and Trustant goes straight to the application list next time. You can change it later from the Configure page — see Configuration.

Once a provider is selected, Trustant runs a short configuration pass: it pulls the models it needs (for Cloud AI), writes the runtime configuration for the Pi coding agent, and finishes with a live “Reply with exactly: OK” probe against the model. Only when that probe succeeds are you let through to the application list — an unusable model would make the whole workbench useless, so a failed check always sends you back to this screen.

Cloud AI

Cloud AI is the fastest way to start. Trustant uses the Ollama server bundled inside it, and that server forwards your requests to Ollama Cloud’s large models, so you get cloud-class quality without installing anything or managing a GPU.

Signing up for Ollama Cloud is free and includes a free allowance you can upgrade later. There is nothing to type into Trustant — no URL, no API key. Instead, you pair this machine with your Ollama account once.

Sign in to Ollama Cloud

The first configuration attempt is expected to fail if you have never paired this device. Trustant detects the authentication failure and shows:

Sign-in required. Trustant has run ollama signin in the background and scraped the pairing URL out of it. Click “Click here to log in to Ollama Cloud” — it opens in a new tab, so this page stays where it is.

If Trustant cannot find a pairing URL, it shows a different message asking you to run ollama signin yourself in a terminal and then click Retry.

Connect the device

The Ollama page asks you to confirm your account and pair this specific machine. The two names shown are the host and the cluster Trustant is running on, so you can tell one installation from another. Click Connect.

Device Connected Successfully. The identity is now stored in the Trustant workspace, so it survives restarts — you only do this once per installation. Close this tab.

Back in Trustant: press Retry

Return to the Trustant tab and click Retry. The configuration pass reruns, now authenticated, and streams its progress:

• Pulling model … / OK: … pulled — one line per model in the recommended Cloud AI set, being fetched into the local Ollama server.
• OK: Pi global configuration written — the coding agent now knows which provider, endpoint and models to use.
• Testing AI model connection… followed by OK: AI model connection successful — the live probe passed.

Click CONTINUE to reach your application list. Setup is done.

Private AI endpoint

Private AI is the option for your own hardware: an Ollama server on your network, a vLLM deployment, a llama.cpp server, or any other service that speaks the OpenAI-compatible API. Nothing about your prompts or code leaves the machines you control, and there is no per-token cost.

The dialog asks for two things:

• Base URL — required. It must end in /v1, the OpenAI-compatible API root, and match the shape http(s)://…/v1. Trustant refuses to save anything else, because a wrong URL would otherwise fail much later in a confusing place.
• API key — optional. Many self-hosted endpoints need none; that is exactly the case this card is named for.

A typical entry is a plain http:// address on your LAN, like an Ollama server listening on port 11434. Because it is http:// and unauthenticated, it saves with no further questions.

If you enter an https:// URL and leave the API key empty, Trustant asks you to confirm — a public HTTPS endpoint almost always requires a key, so an empty one is usually a mistake rather than a choice.

Model discovery

When you click SAVE, Trustant immediately queries the endpoint for its model list rather than trusting the URL blindly. On success it reports how many models it found and tells you the one remaining step. A broken endpoint, or one that returns no models at all, is reported right in the dialog and is never saved.

Choose the default model

Trustant then takes you to the Configure page with a banner: “Choose the Pi model to complete the runtime setup.”

Unlike Cloud AI, a private endpoint carries no catalog telling Trustant which of its models is good at writing code. Picking one automatically would silently select an embedding or a tiny model and fail confusingly later, so Trustant deliberately makes no guess and asks you.

The screen shows:

• Private AI Endpoint — the URL in use. To point at a different endpoint, use Change Provider and pick Private AI again.
• Private AI Models — every model discovered at the endpoint, with its context size and a FOR CODING verdict. Models are marked either Available, or Hidden with the reason — typically “below the recommended 20B minimum”, since small models cannot reliably drive the coding agent. ADD MODEL lets you declare a model the endpoint did not advertise, and REMOVE drops one from the list.
• Pi Model → Default Model — the dropdown where you pick the model the coding assistant will use. Only Available models are offered.

Click SAVE & CONFIGURE to finish. Trustant writes the configuration, runs the “Reply with exactly: OK” probe against the model you chose, and — when it answers — takes you to the application list.

The button is blocked while no default model is selected; Trustant tells you “Please select the default model before saving.”

Next

With a provider configured, continue to the App List to create your first application, or to Configuration to review models and per-application settings.