Get live statistics and analysis of Francisco Moretti's profile on X / Twitter

Hyper-optimized streaming renders! Finally got the perfect combo to achieve this - @aisdk v5 (with a custom store) - Zustand (Selectors) - react-scan (debugging, visualization) This is quite useful, especially with models that answer fast. The usual strategy is to throttle, but now that's no longer necessary. Code is public on the Sparka repo (comments) Gotta thank @nicoalbanese10 for sharing the per-block memoization guide, which takes you quite far. And both @theo and @ksw_arman for showing that this is possible.

No renders while typing with @aisdk 5 Another optimization to improve user experience (faster typing) AI SDK 5 moved the input outside of useChat. You get to decide if it's a state of a ref. 1. Connect your text area with a ref 2. On sendMessage, read the text from that ref

New template: AI SDK + Deep Research Spend a week building the AI SDK version of LangChain's Open Deep Research. I've created a small example repo. Includes: - Requirement gathering phase - Supervisor agent, researcher agents - Type safety with @aisdk - Streamed updates Repo in the comments

Markdown Streaming Memoization v2 v1. Blocks don't re-render v2. Block children don't re-render Useful for - lists: `li` are children of `ol` or `ul`. - format: `bold`, `italics`, `underline` (video). Shipped in Sparka .ai Coming to @aisdk AI Elements / Streamdown soon

All the models on @aisdk AI Gateway Search, filter, compare, explore, chat Vercel AI Gateway continues to add more models, so I've built a way to find the one you want Open source (link below)

Introducing the AI Registry Discover all models in the @aisdk AI Gateway: 🔍 Filter by: - Name - Input/output modalities - Context length - Output size - Providers 📄 View model details ⚖️ Compare model vs. model Open-source, link below!

Loading + Thinking animations Easy thanks to goated libs - Pulse dot (loading) -> Prompt-kit - Shimmer Text (thinking) -> Prompt kit - Reasoning (w/ auto-close) -> AI Elements Only needed to build part grouping for reasoning blocks (gpt-5 thinks a lot) @Ibelick and @haydenbleasel cooked🧑‍🍳 Links in the comments

Optimized streaming rendering with @aisdk 5 Only the new message gets updated on every stream chunk This matters because streaming can send 100+ updates for a message. Ideally, the app can render them all.

AI Gateway might be the most underrated feature in @aisdk 5 - 70+ models with a single API key - Automatic fallbacks - Almost 0 effort to add new models Integrating it with Sparka has been great. It's the perfect fit for a multi-provider chat app. I'm finally able to use the latest models (Quen 3 Coder, Grok 4). And I can use @v0 to generate sweet React code with Shadcn Components.

Show how much context is used (just like cursor) - New `Context` component from AI Elements - Using `tokenlens` from @xn1cklas - Usage data from @aisdk Code in the comments

Migrated to prompt-input from @aisdk AI Elements This input is one of the most complex parts in the whole app. AI elements gives you composable components with great styles. @haydenbleasel cooked as always 🧑‍🍳

Mermaid Streaming is almost 0 cost 1. Rendering: UI component included in Streamdown `npm i streamdown` 2. Backend: Add this to your system prompt `Mermaid: Return diagrams in fenced mermaid code blocks.` Thanks @_ayhansipahi and @haydenbleasel for this gem!