Build Cross-Platform AI Chat Interfaces

MolyKit Skill covers best practices for building cross-platform AI chat interfaces with Makepad using MolyKit - a toolkit providing cross-platform async utilities, ready-to-use chat widgets, AI provider integration via a BotClient trait, an OpenAI-compatible client with SSE streaming, message/bot protocol types, and MCP (Model Context Protocol) support.

Use this when building AI chat interfaces with Makepad, integrating OpenAI or other LLM APIs, implementing cross-platform async for native and WASM targets, creating chat widgets such as messages, prompts, or avatars, or handling SSE streaming responses.

Cross-platform async patterns: PlatformSend (a marker implying Send only on native, implemented by all types on WASM), spawn() (platform-agnostic task spawning - tokio on native, wasm-bindgen-futures on WASM, requiring no Send bound on WASM), AbortOnDropHandle (cancels its future automatically when a widget holding it is dropped), and ThreadToken (stores a non-Send value in thread-local storage, accessed safely via peek/peek_mut across Send boundaries). AI provider integration goes through the BotClient trait (send returns a streamed response, bots lists available bots/models, clone_box clones the trait object), wrapped by BotContext, a shareable Arc<Mutex<...>> handle that loads bots asynchronously and exposes synchronous UI-side accessors. Protocol types include Message (sender EntityId, MessageMetadata, MessageContent), MessageContent (text, reasoning, citations, attachments, tool calls, tool results), and BotId, a globally unique <len>;<id>@<provider> identifier, e.g. BotId::new("gpt-4", "api.openai.com").

Ready-made chat widgets: Slot (runtime content replacement via replace/restore, letting a BotClient supply custom widget content), Avatar (toggles between a text grapheme and an image dependency), PromptInput (send/stop task state, enable/disable, submission detection), and Messages (holds the conversation, scrolls to bottom, tracks bot context). Async work reaches the UI thread via the UiRunner pattern (ui_runner().handle(...), defer_with_redraw), and streaming responses are parsed with parse_sse, which splits an SSE byte stream on blank lines, extracts data: payloads, and filters comments and [DONE] markers before being consumed inside a BotClient::send implementation.

Rust developers building AI chat UIs on Makepad who need the cross-platform async primitives, a concrete BotClient implementation pattern demonstrated against an OpenAI-compatible endpoint, and the chat-specific widget set, guided by eight stated best practices: prefer PlatformSend and spawn() over platform-specific APIs, use AbortOnDropHandle and ThreadToken for cancellation and non-Send WASM state, use Slot for custom content, follow the read()/write() borrow pattern via WidgetRef, update widgets from async contexts through UiRunner::defer_with_redraw, and handle ClientResult's partial-success case where a value and errors can both be present.