Instalar e configurar o SDK para Linux

Integrate Busymate AI in a Linux desktop app using the official Linux SDK distribution 1.0.1: a C library on GTK4 and WebKitGTK 6.0, built with Meson, with GObject Introspection data for Python, GJS, Vala and Rust. Source, the configuration reference and a buildable demo live in the public repository, released as 1.0.1. Identity uses protocol v2 / bridge build 2.0.0; microphone uses protocol v1 / adapter 1.0.0. The release manifest pins the source hashes. Its CONTRACT-v2.md hash matches https://busymate.ai/sdk/v2/2.0.0/SHA256SUMS.

1. Install the pinned version

1. Requirements: GTK 4.10 or later, WebKitGTK 6.0 at 2.42 or later, GLib 2.76 or later, json-glib 1.6 or later and Meson 1.0 or later.
2. Pin the library as a Meson subproject: a subprojects/ wrap file that points at the repository with revision = 1.0.1, then the dependency(…) line the README gives, with example=false and tests=false. Alternatively, install system-wide from tag 1.0.1 with meson setup, meson compile and meson install. Consumers find it through pkg-config and include the SDK header.
3. Create the bridge with the assistant and origins, connect its signals, install it on your web view and create the microphone adapter before webkit_web_view_load_uri. Install both once per web view, before the first load.

Read the versioned installation instructions before wiring callbacks.

2. Load the chat as the window's page

Use the chat URL and assistant slug from your workspace, for example https://YOUR_ASSISTANT.busymate.ai/?channel=desktop_sdk&locale=en. Keep the URL's supported parameters and set channel=desktop_sdk. The chat is the window's top-level page: it uses the normal chat layout and shows your locale from the first frame. Do not use channel=embed, which needs an allowed parent web page and refuses to start as a window, or channel=desktop, which is reserved for the Busymate AI desktop app.

Allow only exact HTTPS origins, without wildcards. The assistant slug adds exactly https://<assistant>.busymate.ai. Add https://busymate.ai when you load a /support/<assistant> URL, and add your mapped support host if you have one.

3. Configure every supported callback

The bridge is configured with construct properties and signals:

Your backend must sign the supplied nonce with ES256, a unique jti and an expiry within 120 seconds. Never ship signing keys, cache launch tokens or log tokens. See backend signing and the complete configuration reference.

4. Wire the OS microphone callback

Call the bridge's account_changed function after login, logout and account switches. When the web process crashes, the bridge reloads the chat and sends restored.

Unsandboxed apps have no per-app runtime microphone prompt; capture goes through WebKitGTK to PipeWire or PulseAudio. Under Flatpak, add the --socket=pulseaudio finish-arg. Under Snap, connect audio-record. The adapter grants WebKit capture for audio only, for an allowed origin, after a tap. The optional request-permission signal lets you show your own consent dialog.

A mic or voice tap sends busymate.microphone.v1.request with id and source. The SDK obtains OS permission and returns { id, granted }. Chat waits for that reply before capture or voice token minting. Denial and timeout fail closed, and cancellation prevents a delayed grant from starting a cancelled session. Opening the chat never asks. See microphone lifecycle and troubleshooting.

5. Build and test the demo

Run meson setup _build && meson compile -C _build, then run the example binary under _build/example-app/. Its Settings dialog changes the chat URL, locale, origins, guest or identified mint, microphone and close callbacks. Follow the example-app build instructions. Guest chat needs no fake identity, and authenticated tests use your own backend. The event log helps with integration troubleshooting without exposing credentials.

Knowledge, tools, handoff, branding and voice availability are workspace features. Locale and appearance follow the supported hosted chat parameters. The SDK does not invent native flags for these services.

Verify

1. Build the example app and verify guest chat, loaded with channel=desktop_sdk.
2. Load ?channel=desktop_sdk&locale=de: the first frame is in German and the chat fills the window with no extra title-bar inset.
3. Use your backend to test sign-in, logout and account switching without history leakage.
4. On a real machine, test the first mic/voice grant, denial, retry after changing the OS setting, and cancellation while the prompt is open.
5. Test external links, close/action callbacks, window re-activation and renderer recovery.