Changelog

The AI Components library grew from a handful of chat pieces to a full kit for agent apps. Every component reads AI SDK message or tool parts directly, installs with the shadcn CLI, and now has its own page under /ai-components.

New AI Components

Inputs

• Mention Menu: an @ menu for files, docs, people, and tools. Mentions delete as one piece and send as data parts.
• Slash Command Menu: a / menu with recent commands first and the remaining arguments shown in the field.
• Starter Prompts: the first screen of a chat, as a list or a row of cards, chosen by connected tools and recent work.
• Attachments Tray: file and image chips with upload progress, retry, a media viewer, and drop or paste to add.
• Model Picker: choose a model and a reasoning level, with context window and relative cost per model.
• Generative Form: a form that fills itself in from a streaming object, then lets the user correct it and send.
• Line Items: rows that stream in for invoices and orders, with editable amounts, tax, and a running total.
• Point and Ask: drop a pin or draw a box on an image, then ask about that spot.
• Voice Session: voice mode in one card, with a live wave, transcript, mute, interrupt, and end.

Responses

• AI Answer: a grounded answer with numbered citations that light up their source cards, plus follow-ups and feedback.
• Source Viewer: opens a document at the cited passage and warns when a quote is not in the text.
• Artifact Panel: a side panel for documents, code, and charts, with version history and live streaming.
• Live Preview: runs model-written HTML in a frame and sends runtime errors back to the model in one click.
• Model Compare: one prompt answered by two or three models side by side, with time and cost on each.
• Message Actions: copy, edit and resend, retry on another model, ratings, and a version switcher.
• Selection Actions: select text in an answer to ask about it, explain it, quote it, or copy it.

Tool Results

• Tool Call Card: a compact row per tool call that expands to highlighted input and output.
• Data Table Result: query rows as a sortable table with typed columns and copy as CSV.
• Chart Result: a tool result drawn as a line, area, column, bar, or pie chart, with no chart library.
• Image Results: generated images that fade in as they arrive, with a large view, download, and reuse.

Approvals

• Agent Plan: a plan the user edits before approving, which then becomes the run log.
• Edit Before Run: correct a tool call's arguments before it runs, and get back what changed.
• Decision Inbox: one place for the approvals background agents are waiting on.
• Guardrail Notice: one sentence on what a guardrail caught, with a way to ask for review.
• Schedule Card: a recurring job with a live countdown, past runs to scrub, and pause or run now.
• Prose Diff: review an AI rewrite in place and accept or reject each suggestion by click or key.

Feedback

• Thinking State: reasoning, tool calls, and sources as a step list or a one-line pill.
• Agent Run Tree: parallel subagents as a live tree with step, time, tokens, and status per agent.
• Research Job: a long-running task the user can walk away from, with progress, sources, and findings so far.
• Task Tray: a small pill for background agents that says who needs you, what is running, or what finished.
• Browser Session: watch an agent drive a browser, review each step, and take over.
• Context Window: a map of the token budget by system prompt, tools, files, conversation, and memory.
• Run Receipt: what a run cost and where it went, priced per line with cache savings.
• Eval Scorecard: cases passing, averages per criterion, and what was fixed or regressed since the last run.
• Review Queue: grade real outputs from the keyboard and get back labeled rows for an eval set.
• Memory Manager: what the agent remembers, one sentence per memory, with edit, delete, and undo.
• Knowledge Sources: the files and sites an agent can search, with indexing progress and citation counts.
• MCP Server List: connection status, a switch per server and per tool, and reconnect for the ones that dropped.

Code

• File Diff: a unified diff with word-level highlights, collapsed hunks, and optional accept and reject.
• Terminal Result: a command the agent ran, with exit status, a live timer, and output that keeps its ANSI colors.
• Test Run Result: only what failed, with expected and received values diffed and a Fix action.

Navigation

• Session List: chat history for a sidebar with search, pinning, inline rename, and a confirmed delete.
• Session Tabs: open chats as draggable tabs with a running or unread mark on each.
• Message Branches: a version pager that opens a tree of every edit and retry of a message.

Two more Jev playgrounds for matching and grouping. Code narrows the candidates, Jev answers one boolean per pair, and application code makes the final call.

New Patterns Added

Jev Entity Resolution

Resolve messy card statement descriptors to the merchants behind them. Code blocks and vetoes candidate pairs first, and unsure matches wait for review. Available at /patterns/jev-entity-resolution.

Jev Semantic Clustering

Find themes in a pile of comments with no labels up front. Embeddings pick the likely pairs and code does the grouping. Available at /patterns/jev-semantic-clustering.

Two Pro patterns and one free pattern: gated long-term memory, a grid-first data agent, and image generation with xAI.

New Patterns Added

Jev Memory

Long-term agent memory with three Jev gates. Keep durable facts, recall what fits a token budget, and evict what went stale. Nothing is rewritten. Available at /patterns/jev-memory.

Spreadsheet Agent

The grid is the main UI, with a floating copilot beside it. The agent emits clean and transform operations that apply in the browser, charts the results, and exports a multi-sheet .xlsx. Available at /patterns/spreadsheet-agent.

Generate Image (Grok Imagine)

Generate images from a text prompt with xAI Grok Imagine through the Vercel AI Gateway. Available at /patterns/basics-generate-image-grok.

Three more Jev playgrounds: a FAQ that grows from grounded asks, a live-search leave-one-out weight demo, and a per-PR eval suite.

New Patterns Added

Jev Infinite FAQ

A closed accordion of product answers, plus a live ask field that looks like another title. Jev opens a canned row at 0.70, or writes a short passage-only answer that becomes a new row. Available at /patterns/jev-infinite-faq.

Jev Tool Permissions

Describe a task and the tools the agent won't need drop out before the model sees them. Allow and deny rules run first without a model call, every other tool gets one Jev relevance boolean in a single call, and anything under 0.20 is dropped. Ships a prepareStep helper for streamText. Available at /patterns/jev-tool-permissions.

Jev Web Context

Exa finds live evidence, Jev answers the same question with and without it, and leave-one-out weights in application code show what each source changed. Available at /patterns/jev-web-context.

Jev Evals

A support-agent suite run with jev-evals. One Jev call per case answers every rubric, the run reports what it actually cost, and a baseline diff against main flags fractional regressions. Available at /patterns/jev-evals.

Filter or sort a closed invoice table. A phrase that only names a column never leaves the app.

New Pattern

Jev Table Query

Filter dims rows under a 0.70 boolean floor. Sort reorders by a four-level score. Amount, date, and vendor sort in the browser. Available at /patterns/jev-table-query.

A closed inbox, four tags, and stamps that land only when they clear the floor.

New Pattern

Jev Email Tags

One Jev boolean per message and tag. Receipt, Reply, Waiting, and News stamp the row at 0.70 or above. Mail that earns nothing stays bare. Available at /patterns/jev-email-tags.

Paste a public URL. Firecrawl fetches the page, application code measures title, meta, canonical, and noindex, and Jev scores whether the answer block is worth citing.

New Pattern

Jev AEO Audit

Four boolean checks, a query-fit score, and an intent choice. Cite-ready, human review, or a rewrite queue is decided in application code. Jev does not rewrite the page. Available at /patterns/jev-aeo-audit.

Seven more Pro patterns: six Jev evaluation playgrounds that keep thresholds in application code, plus a code-mode tool-calling compare.

New Patterns Added

Jev Output Verifier

Write an answer, then show it only if Jev is at least 90% sure it answers the question. If not, rewrite once. Available at /patterns/jev-output-verifier.

Jev Loop Breaker

Look up facts one step at a time. Stop when Jev is at least 90% sure the notes answer the task, or after three lookups. Available at /patterns/jev-loop-breaker.

Jev Best-of-N Judge

Three prepared replies sit on the table. A Jev choice lifts the winner only when its probability clears 0.70. Available at /patterns/jev-best-of-n-judge.

Jev Sub-agent Delegation

Pick one specialist with a Jev choice, then run only that agent when confidence and selected probability both clear their floors. Available at /patterns/jev-sub-agent-delegation.

Jev Search Filter

Filter a closed catalog with one Jev boolean per item. Matches rise out of the pile when their probability clears 0.70. Available at /patterns/jev-search-filter.

Jev JSON Render

Compose an account settings form by letting Jev select prepared json-render candidates—no invented specs. Available at /patterns/jev-json-render.

Code Mode Compare

See if stock covers an open order. The model writes one program that calls both lookups and subtracts inside code_mode. Available at /patterns/code-mode-compare.

A new AI Component for the moment an agent has to stop and ask a person.

New AI Component

Agent Gate

Paged questions, plan approval, plugin connect, permission, eval scorecard, secret access, handoff, and scheduled runs — one gate, eight variants. Browse it on /ai-components.

We've shipped six new Pro patterns: three browser-side retrieval tools (image ingest, image search, and chunk-strategy compare) plus three Jev evaluation playgrounds that keep thresholds and scoring in application code.

New Patterns Added

Local Image Ingest

Caption a curated image with a vision model, embed the caption, and export a portable aisdkagents.local-image-index JSON file—without a vector database. Pair it with Local Image Search to query the index. Available at /patterns/local-image-ingest.

Local Image Search

Import a portable image index (or use the shipped sample), embed one query, and rank captions in the browser with cosine similarity. Images below the floor stay visible but dimmed. Available at /patterns/local-image-search.

Chunk Strategy Compare

One document, chunked three ways (small windows, large windows, paragraph-window). Embed a query and see which strategy retrieves best—without a vector database. Available at /patterns/local-rag-chunk-compare.

Jev Action Gate

Gate a proposed action with one Jev boolean probability. Application code owns the auto-run threshold—the model only answers whether the action looks safe to run unattended. Available at /patterns/jev-action-gate.

Jev Confidence Router

Pick one of N queues with a Jev choice question, then assign only when both confidence and the selected option's probability clear their floors. Ambiguous cases go to review. Available at /patterns/jev-confidence-router.

Jev Composite Scorecard

Ask three atomic Jev score questions against a pull request description, then combine the answers with weights in application code. The model never ranks the whole request in one question. Available at /patterns/jev-composite-scorecard.

We've shipped two new Pro patterns—browser-side RAG chat and Supermemory long-term memory tools—plus a new Prompt Input AI Component for chat composers.

New Patterns Added

Local RAG Chat

Load a portable Local RAG index, rank chunks in the browser with cosine similarity, and stream a cited answer—without a vector database. Pair it with Local RAG Ingest to build the index, then chat here. Available at /patterns/local-rag-chat.

Supermemory Tools

Save and recall durable facts with Supermemory's AI SDK tool pack (addMemory, searchMemories) in a simple Remember / Recall playground. Available at /patterns/tool-memory-supermemory.

New AI Component

Prompt Input

Expandable chat composer with @ sources, / skills, attachments, model picker, and rounded or pill variants. Browse it on /ai-components.

We've added Hono support so you can export a pattern's server logic as a standalone API service—run it on Node.js and call it from any frontend over HTTP.

What's New

Supported patterns now offer Download → Hono API server and Copy for a coding agent → Hono API server alongside the existing Next.js download and shadcn install flows. Each export includes the pattern's AI and API routes, a generated Hono entry point, runtime dependencies, .env.example, a pattern-specific README, and a /healthz endpoint—without the React UI or Next.js app shell.

Hono exports target Hono on Node.js (>=24). Docs cover exporting, connecting a frontend, deploying with production CORS, and compatibility limits at /docs/blocks/hono.

We've expanded Premium skills into a full Plan → Build → Evaluate → Optimize lifecycle, with a new skills catalog at /skills.

New Skills Added

Plan AI SDK Agent

Inspects your repository and writes an implementation-ready Markdown blueprint covering architecture, tools, state, security, tests, and build order—without changing application code. Available at /skills/plan-ai-sdk-agent.

Build AI SDK Agent Evals

Adds datasets, deterministic tests, model-graded scorers, and local or CI commands around an existing AI SDK agent. Live model runs are opt-in. Available at /skills/build-ai-sdk-agent-evals.

Optimize AI SDK Agent

Baselines an existing agent, tests focused changes, reruns the same checks, and keeps only improvements supported by comparable results. Available at /skills/optimize-ai-sdk-agent.

Skills Catalog

Browse all four Premium skills—including the existing AISDK Agents build skill—on one page with install commands for Cursor, Claude Code, Codex, and other skill-aware tools.

We've shipped a filterable Patterns catalog so you can browse free and Pro AI SDK patterns by outcome, API, and use case before you install.

What's New

Open /patterns to search and filter the library, preview requirements and access, then jump into pattern detail with clearer setup and export actions—including Hono when a pattern is verified.

We've launched AI Components—a registry of high-quality animated UI components for AI apps, inspired by interfaces from OpenAI, Claude, Cursor, and more.

What's New

Browse interactive, tool-rendered component previews and install registry-ready source with the shadcn CLI. Includes chat UI pieces for recipes, structured output, maps and itineraries, image loaders, input surfaces, and marketing-ready AI Components—available at /ai-components (Pro subscribers get the full set at /pro/ai-components).

We've shipped the AISDK Agents skill—a Premium agent playbook you can install into skill-aware IDEs like Cursor and Claude Code.

What's New

Give your AI assistant the production agent architecture patterns from AI SDK Agents: pattern decision trees, current AI SDK APIs (ToolLoopAgent, streaming, structured output), routing and orchestration, tools and repair, human-in-the-loop, artifacts, and anti-pattern checks. Install with:

bunnpmpnpmyarn
pnpm dlx skills add https://aisdkagents.com/skill/<token> --skill aisdkagents
Copy

Learn more at /skills/aisdkagents.

We've shipped four new Pro patterns: three download-first document artifacts (Word, PDF, Excel) plus a streaming GeoJSON map artifact.

New Patterns Added

Agent DOCX Artifact

Download-first Word artifact pattern. The agent streams Markdown through a tool call, renders a document preview, and exports a real .docx file.

Agent PDF Artifact

Download-first PDF artifact pattern. The agent streams Markdown through a tool call, renders a live PDF preview, and exports a real .pdf file.

Agent XLSX Artifact

Download-first Excel artifact pattern. The agent streams CSV through a tool call, renders a spreadsheet preview, and exports a real .xlsx file with SheetJS.

Agent Map Artifact

Template-style GeoJSON map artifact with data-mapDelta streaming, MapLibre preview and JSON source editor, GeoJSON export, in-memory versions, and quick prompts—no database persistence.

We've migrated the entire pattern library to AI SDK v7.

Platform Update

Every Pattern and Template now runs on AI SDK 7 APIs—including ToolLoopAgent, stabilized structured output, updated streaming and tool lifecycles, and the v7 naming conventions (instructions instead of system, prepareStep, and related callback renames). Existing Patterns stay installable and up to date with the current SDK surface.

We've added two Pro patterns that load filesystem skills into a ToolLoopAgent with scoped readFile access—so agents pull SKILL.md instructions and reference docs on demand.

New Patterns Added

Just Bash Skills Agent

Minimal ToolLoopAgent chat demo: on-call-style skills (incident triage + stakeholder comms) on disk, skill tool loads SKILL.md + paths, readFile pulls reference checklists and templates from the sandbox.

Marketing Agent with Skills + readFile

Chat UI wired to a ToolLoopAgent that loads filesystem skills and exposes readFile on the skill sandbox—demo includes ad-creative, copywriting, and paid-ads skill packs for campaign workflows.

We've migrated our shadcn UI primitives from Radix UI to Base UI.

Platform Update

Component wrappers across the site and registry now use @base-ui/react instead of @radix-ui/*—dialogs, menus, popovers, toggles, and the rest of the primitive set. Installs and demos stay aligned with the current shadcn Base UI styles while keeping the same composition patterns you already use.

We've shipped six new Pro patterns covering ToolLoopAgent request configuration, loop control, context injection, and prompt caching through Vercel AI Gateway.

New Patterns Added

Call Options: prepareCall

Dynamically configure one ToolLoopAgent per request with callOptionsSchema and prepareCall. A customer-support agent rewrites instructions, gates tools, tunes reasoning effort, and injects RAG context based on account tier and urgency.

Loop Control: stopWhen

Demonstrate ToolLoopAgent loop control with stopWhen. Covers stepCountIs, hasToolCall, combined conditions, and a custom StopCondition with step timeline visualization.

PrepareStep: Trim Message History

Demonstrate ToolLoopAgent prepareStep for trimming long message history. Pre-seeded 24-message conversation shows context compression on the first run with per-step trim telemetry.

Tool Context: experimental_context

Inject request-scoped context (user identity, database handles, session) into ToolLoopAgent tools via experimental_context, so the model never supplies sensitive identity in tool inputs.

OpenAI Prompt Caching (AI Gateway)

Lightweight ToolLoopAgent pattern for OpenAI prompt caching through Vercel AI Gateway: stable system instructions, promptCacheKey, and per-turn cachedPromptTokens metrics.

Anthropic Cache Control (AI Gateway)

Lightweight ToolLoopAgent pattern for Anthropic ephemeral cache control through Vercel AI Gateway: prepareCall injects a cached system breakpoint and reports cache read/write tokens.

We've added a Pro scraper that converts any webpage to clean Markdown via Cloudflare's markdown.new.

New Patterns Added

Cloudflare Markdown.new Web Scraper

Convert any webpage to markdown using Cloudflare's markdown.new. Clean, structured output for AI pipelines and content processing.

We've shipped five new Pro patterns: three template-style agent artifacts (Mermaid diagrams, slide decks, and Excalidraw whiteboards) with streaming UI data parts, plus two Firecrawl-powered brand image workflows for mood boards and brand-standards manuals.

New Patterns Added

Agent Mermaid Artifact

Template-style Mermaid artifact with data-mermaidDelta streaming, diagram preview and source editing, SVG and PNG export, in-memory versions, and quick prompts—no database persistence.

Agent Slides Artifact

Template-style slide deck artifact with data-slidesDelta streaming, a sortable slide list and preview, Markdown export, in-memory versions, and quick prompts—no database persistence.

Agent Canvas Draw Artifact

Excalidraw whiteboard artifact with data-whiteboardDelta streaming, an uncontrolled canvas, SVG, PNG, and .excalidraw export, scene templates (architecture, user flow, org chart), and in-memory versions—no database persistence.

Brand Mood Board Generator

Paste a website URL to extract brand signals with Firecrawl, generate a structured mood-scene brief for abstract and environmental imagery, auto-create four brand-world images, and open any result in a local edit flow.

Brand Guidelines Image Generator

Paste a website URL to extract brand signals with Firecrawl, generate a structured brand-standards brief, render manual-style spreads (color, typography, grid, photography, and more), and refine any image with edit instructions.

We've shipped seven new Pro patterns: three template-style agent artifacts (sheet, text, and code) with streaming UI data parts, one JSON Render Pattern for AI-generated data tables, and three client-side export utilities for markdown and tabular data.

New Patterns Added

JSON Render Data Table

Generate sortable, filterable tables from natural language with json-render. Streams specs into an interactive DataTable with toolbar satellites (search, filters, pagination, bulk actions).

Agent Sheet Artifact

Template-style spreadsheet artifact with data-sheetDelta streaming, a CSV grid editor (react-data-grid), in-memory versions, and quick prompts—no database persistence.

Agent Text Artifact

Template-style document artifact with data-textDelta streaming, a ProseMirror Markdown editor, in-memory versions, and quick prompts—no database persistence.

Agent Code Artifact

Template-style code artifact with data-codeDelta streaming, a CodeMirror Python editor, optional Pyodide execution, in-memory versions, and quick prompts—no database persistence.

Markdown multi-format file export

Parse markdown into H1 sections with unified and remark, then export normalized content as JSON, CSV, Markdown, HTML, PDF, or plain text in the browser (no API keys).

CSV / TSV ↔ JSON tabular export

Paste CSV, TSV, or JSON arrays; validate rectangular rows; preview in a table; export JSON, CSV, TSV, or GitHub-flavored Markdown tables in the browser.

XLSX → CSV export

Load Excel .xlsx or .xls workbooks in the browser with SheetJS, pick a sheet, preview rows, and download CSV—with optional UTF-8 BOM and multi-sheet export.

We've added a Pro Pattern that turns short UI requests into production-grade design briefs.

New Patterns Added

Prompt UI Improver

Transform vague UI prompts into structured design briefs with strong visual, interaction, and implementation detail—ready to hand to design and engineering workflows.

We've shipped a new Google image generation Pattern, rolled out broad UI refreshes across basics and Pro demos, and synchronized registry metadata and previews across internal/public surfaces.

New Patterns Added

Generate Image (Google Gemini)

Added a new basics Pattern for image generation with Google Gemini, including a dedicated API route, complete demo UI, dark/light preview assets, and full registry wiring for style bundles and block indexing.

Registry & Metadata Updates

• Added basics-generate-image-google to block registries and generated style manifests.
• Updated registry indexes and category wiring across apps/v4/registry and published style registries.
• Renamed multiple image-related Pro Pattern labels and descriptions from "Gemini Flash" to "Nano Banana Flash" for clearer naming consistency.
• Refreshed pro-registry/internal registry JSON metadata to match current block definitions.

Basics & Pro UI Refresh

Basics Forms and Tools

Refined core basics demos and tool UIs for consistency and improved UX across:

• image generation (fal, openai)
• object/text generation and streaming demos
• Cheerio, Jina, and Markdown-new scraper demos/tool panels
• chat-claude and marketing code block examples

Pro Patterns and Structured Output Demos

Updated major Pro experiences and shared displays, including:

• structured output array/choice demos and output renderers
• AI Elements tool demo
• scraper Pro Patterns (Cheerio/Jina/Markdown-new) app pages, forms, and result displays
• json-render-remotion and json-render-threejs demo components

Brand Image & Firecrawl Workflow Improvements

• Expanded and refined example-brand-image-generation workflow logic, schema handling, and rate-limit integration.
• Updated firecrawl-brand-pdf countdown/empty state/demo components for improved interaction flow.
• Refreshed generated preview screenshots for brand image generation examples.

Platform & Runtime Support Updates

• Updated pattern client and view API route behavior for registry/view interactions.
• Refined supporting utilities and script flows used by registry capture/distribution.
• Synchronized root registry artifacts (apps/v4/registry.json, style registry JSONs, and block index outputs) with the current block set.

We've shipped three new Pro Patterns and a large component refresh across the JSON Render demos and orchestration examples.

New Patterns Added

Brand Product Placement Generator

Paste a website URL to extract brand context with Firecrawl, generate a product photography brief, and create multiple placement images with local edit flows. Includes the new gemini-flash-image, dropzone, countdown, block-header, and icon components.

JSON Render Email

Generate React Email templates from prompts with live preview plus copyable HTML and plain-text output. Includes a dedicated empty state and streaming demo component built for email-safe layouts.

JSON Render Three.js

Generate interactive 3D scenes from natural language with JSON Render and Three.js. Includes a new Three.js demo component, API route integration, and a dedicated empty state.

Component Updates

Example Brand Image Generation

Added and integrated new UI components for the full placement workflow:

• block-header
• countdown
• dropzone
• gemini-flash-image
• icons

JSON Render Component Refresh

Updated Pro Pattern components across the JSON Render stack:

• json-render-email/components/empty-state
• json-render-email/components/json-render-demo
• json-render-generate/components/json-render-demo
• json-render-image/components/empty-state
• json-render-image/components/json-render-image-demo
• json-render-pdf/components/empty-state
• json-render-pdf/components/json-render-pdf-demo
• json-render-remotion/components/json-render-remotion-demo
• json-render-shadcn/components/json-render-demo
• json-render-threejs/components/empty-state
• json-render-threejs/components/json-render-demo

Orchestrator Demo Update

Refreshed the sub-agent-orchestrator/components/orchestrator-demo component for the latest Pro Pattern behavior and UX patterns.

Pro Pattern Index Updates

• Added json-render-email to POPULAR_PRO_BLOCKS.
• Updated preview flags for example-brand-image-generation, json-render-email, and json-render-threejs.
• Set firecrawl-landing-page-teardown-pdf preview availability to locked mode.

We've added three new Patterns: an SEO audit agent with Firecrawl, a full-screen brand strategy levee interface, a user research levee flow, plus a new platform feature for copying files into AI tools.

Copy Skill Files

Each Pattern now highlights the files that matter most—the core AI logic, agent implementations, and tool definitions. Use Copy for a coding agent in the Pattern viewer toolbar to copy those files with AI instructions; paste them into Cursor, Claude, or any AI tool so it gets the code and context needed to help you iterate. Available on all free Patterns; Pro and Premium subscribers get access on every Pattern.

New Patterns Added

SEO Audit

Audit any website for on-page SEO issues using Firecrawl. Scrapes a URL, then renders a full-screen report analyzing meta tags, content structure, links, keywords, and schema markup with exportable reports.

Levee Brand Strategy

A full-screen AI levee interface where each stage is an adaptive interview with lock-based progression and dynamic tool-driven prompts.

Levee User Research

A full-screen AI levee interface that guides users through building a complete user research plan and interview guide, with lock-based progression between stages.

We've added seven new agent Patterns for business intelligence, accessibility auditing, data analysis, and advanced planning workflows, plus a new ecommerce multi-agent Template.

New Templates Added

Ecommerce Multi-Agent

A full-featured ecommerce AI assistant with multi-agent orchestration, product browsing, shopping cart management, and real-time artifact streaming. Features intelligent triage agent routing to specialized agents (Products, Account, General), comprehensive product discovery tools, shopping cart management, and rich artifact streaming with interactive UI components.

New Patterns Added

Brand Design System Extractor

Extract brand design systems from any website using Firecrawl, then use AI tools to export tokens, generate color palettes, check accessibility, and analyze brand personality.

Competitor Intelligence Agent

Enter a competitor URL to get comprehensive competitive intelligence. Uses Firecrawl Map + Parallel Scrape for fast extraction (~3-5s). Analyzes positioning, pricing, features, tech stack, and generates sales battle cards.

Data Analysis Agent

Upload CSV or JSON data and get AI-powered analysis with interactive visualizations. Detect patterns, correlations, outliers, and generate insights with confidence scores. Export comprehensive reports.

Accessibility Audit Agent

Audit any website for WCAG 2.1 accessibility compliance using Firecrawl, then use AI tools to analyze issues by severity, check color contrast, validate heading structure, generate alt text, and create prioritized remediation plans with visual tool outputs.

Reddit Idea Validation Agent

Validate startup ideas with a two-phase workflow: discover relevant subreddits, then analyze pain points, existing solutions, willingness to pay, demand trends, and early adopters. Get a go/no-go scorecard with actionable insights.

Agent HIL Plan

Uses ToolLoopAgent pattern to gather requirements and generate plans. Generates markdown plans with structured todo lists for organized task execution.

Marketing Plan Agent

Two-phase marketing agent with brand context: Stage 0 extracts brand info from URL or manual input, planning phase gathers requirements and creates a plan with todos, implementation phase executes the plan using web scraping and research tools. Uses callOptionsSchema + prepareCall for dynamic agent configuration.

We've added a hyper minimal sub agent template for building multi-agent systems with AI SDK Tools.

New Templates Added

Sub Agent Starter

A hyper minimal sub agent template featuring multi-agent orchestration, tool usage, and artifact streaming with AI SDK Tools. Includes intelligent triage agent routing to specialized agents (billing, technical, general), support tools for customer lookup, order status, and ticket creation, plus structured artifact streaming for order summaries and support tickets.

We've added a comprehensive AI agent platform template with RAG capabilities and multi-tenant support.

New Templates Added

AI SDK Agent Platform

Build and deploy RAG agents with this production-ready platform. Features multi-model AI support (50+ models from OpenAI, Anthropic, Google), document upload and knowledge base creation with pgvector embeddings, Vercel Workflows for async processing, multi-tenant organizations with role-based permissions, agent management with 7-step creation wizard, and built-in tools including Exa search and Cheerio web scraping.

We've added five powerful Workflow DevKit workflow patterns demonstrating durable workflow execution, plus advanced agent orchestration and form generation capabilities.

New Patterns Added

Evaluator Workflow Pattern

Demonstrates an evaluator-optimizer workflow pattern using Workflow DevKit. Features iterative improvement through AI evaluation and optimization loops.

Orchestrator-Worker Workflow Pattern

Demonstrates an orchestrator-worker workflow pattern using Workflow DevKit. Features parallel execution of specialized workers for different change types (create, modify, delete).

Parallel Review Workflow Pattern

Demonstrates a parallel review workflow pattern using Workflow DevKit. Features concurrent code reviews (security, performance, maintainability) with result aggregation.

Routing Workflow Pattern

Demonstrates a routing workflow pattern using Workflow DevKit. Features query classification and conditional routing to specialized handlers based on type and complexity.

Sequential Workflow Pattern

Demonstrates a sequential workflow pattern using Workflow DevKit. Features step-by-step execution with quality gates and conditional regeneration.

AI Form Generator

Generate Zod schemas from natural language descriptions using AI. Automatically creates form schemas that render interactive forms with real-time validation.

Sub-Agent Orchestrator

Custom Agent implementation demonstrating the Agent interface abstraction with an orchestrator that routes queries to specialized sub-agents (research, analysis, support). Shows options passing and structured outputs.

Agent Chain with Structured I/O

Sequential three-agent chain demonstrating structured outputs flowing between agents. Agent 1 gathers research, Agent 2 expands research, Agent 3 synthesizes final answer. Uses Exa tools and AI SDK 6's stabilized structured output support.

@exalabs webSearch Tool

Search with Exa AI SDK's webSearch tool and Anthropic Claude Sonnet 4.5. Uses @exalabs/ai-sdk for seamless integration.

We've added two new agent API Patterns demonstrating structured agent outputs with constrained choices and arrays for type-safe, predictable agent responses, plus a production-ready multi-tenant starter Template.

New Patterns Added

Structured Agent Output: Output.choice

Demonstrate structured agent outputs using ToolLoopAgent with Output.choice. Constrain agent outputs to predefined choices for predictable, type-safe responses.

Structured Agent Output: Output.array

Demonstrate structured agent outputs using ToolLoopAgent with Output.array. Returns structured array output with element schema for predictable, type-safe responses.

New Templates Added

Multi Tenant Better Auth Chat

Production-ready starter for building multi-tenant, organization-based Next.js applications. Features Better Auth for authentication, Drizzle ORM for PostgreSQL, Stripe for subscriptions, and complete organization/team management with roles and permissions. Perfect for SaaS applications requiring multi-tenant architecture.

We've added two new Pro Patterns: a comprehensive Chatbase clone with advanced chat capabilities and a Gemini-powered image merge tool for creative workflows.

New Patterns Added

Chat Base Clone

A comprehensive Chatbase clone demo with widget interface, model selection, web search integration, and human-in-the-loop tool approval workflows. Features a fully functional chat widget with multiple AI model support and interactive tool management.

AI SDK Gemini Flash Image Merge

Generate, edit, and merge images using Google's Gemini 2.5 Flash. Create new images from text, edit existing ones with natural language, or merge two images together. Includes version history and image comparison.

We've added five new SDK API tool Patterns that demonstrate advanced tool execution patterns and capabilities. These Patterns showcase dynamic tool creation, context passing, input lifecycle tracking, preliminary results, and tool call repair.

New Patterns Added

Tool API Context

Pass arbitrary context from generateText/streamText to tool execution via toolsContext. Access context in tool execute functions for dynamic behavior and state management across tool boundaries.

Tool API Dynamic Tool

Create dynamic tools with runtime-generated schemas using dynamicTool. Demonstrates unit conversion tools where schema adapts based on conversion type (temperature, length, weight, currency) and supports dynamic model selection.

Tool API Input Lifecycle Hooks

Explore tool input lifecycle hooks: onInputStart, onInputDelta, and onInputAvailable. See how to track tool input generation in real-time during streaming for enhanced user feedback and progress tracking.

Tool API Preliminary Tool Results

Return preliminary tool results using async iterables. Stream status updates during tool execution before the final result is available, enabling real-time progress indicators and intermediate feedback.

Tool API Tool Call Repair

Repair invalid tool calls using experimental_repairToolCall. Fix schema validation errors without requiring additional steps that pollute message history, providing seamless error recovery and improved user experience.

We've added a new AI Elements component for displaying AI-generated execution plans with streaming support.

New Patterns Added

Plan

A collapsible plan component for displaying AI-generated execution plans with streaming support and shimmer animations.

We've expanded our basics collection with multi-provider support for core AI capabilities and added powerful new tools for web scraping and search. Now supporting OpenAI, Anthropic, Google, FAL, and ElevenLabs across various AI operations.

New Patterns Added

Generate Image (FAL)

Image generation using FAL's API with fast generation times and efficient processing.

Generate Image (OpenAI)

Image generation using OpenAI's DALL-E models with advanced prompt capabilities and style controls.

Generate Object (Claude)

Structured object generation using Anthropic's Claude models with Zod schema validation and type safety.

Generate Object (Google)

Structured object generation using Google's Gemini models with intelligent schema interpretation.

Generate Object (OpenAI)

Structured object generation using OpenAI's GPT models with structured output capabilities.

Generate Speech (ElevenLabs)

Text-to-speech using ElevenLabs' voice synthesis API with natural voice generation and emotion control.

Generate Speech (OpenAI)

Text-to-speech using OpenAI's TTS models with multiple voice options and adjustable speech parameters.

Transcribe Audio (ElevenLabs)

Audio transcription using ElevenLabs' Scribe models with accurate speech-to-text conversion and multi-language support.

Transcribe Audio (OpenAI)

Audio transcription using OpenAI's Whisper models with segment timestamps and language detection.

Tool Approval Confirmation

Demonstrate tool execution with user confirmation. Includes approval workflows, confirmation UI, and state management for safe AI operations.

Tool: Scrape with Cheerio

Extract content from websites using Cheerio for HTML parsing with link discovery and rate limiting.

Tool: Scrape with Jina

Advanced web scraping using Jina AI's reader API with JavaScript rendering and proxy rotation.

Tool: Search with Exa

Semantic web search using Exa's API with real-time crawling and relevance scoring.

Tool: Search with Firecrawl

Web search and content extraction using Firecrawl's API with URL processing and markdown conversion.

Search - Exa AI (robust)

Search the web using Exa AI's API and real-time crawling. Includes URL search, content extraction, and semantic search tools.

Search - Firecrawl (robust)

Search the web using Firecrawl's API and real-time crawling. Includes URL search, content extraction, Wikipedia, Reddit, and news search tools.

Scrape - Cheerio (lightweight)

Extract content from websites using Cheerio for HTML parsing. Lightweight scraping solution with link discovery and rate limiting.

Scrape - Jina AI (advanced)

Advanced web scraping using Jina AI's reader API with JavaScript rendering and proxy rotation. Handles rate limiting and bot detection.

We've added four foundational AI Patterns for getting started with AI integration, plus three advanced human-in-the-loop Patterns for interactive workflows.

New Patterns Added

Generate Text

A clean, simple AI prompt interface with text generation capabilities. Features a streamlined input form with character counting, loading states, and response rendering using the generateText API.

Stream Text

A clean, simple AI prompt interface with streaming text generation capabilities. Features a streamlined input form with character counting, loading states, and real-time response rendering using AI Elements and the useCompletion hook.

Generate Object

A clean, simple AI prompt interface with structured object generation capabilities. Features a streamlined input form with character counting, loading states, and structured data rendering using the generateObject API and Zod schemas.

Stream Object

A clean, simple AI prompt interface with structured object generation capabilities. Features a streamlined input form with character counting, loading states, and real-time structured data rendering using AI Elements and the experimental_useObject hook.

HIL Needs Approval

Demonstrate tool approval with async generators. Includes real-time feedback, approval workflows, and interactive management.

HIL Inquire Multiple Choice

Gather information through AI-driven multiple choice questions. Includes approval workflows and real-time rendering.

HIL Inquire Text Input

Gather detailed information through AI-driven text input questions. Includes approval workflows and real-time rendering.

We've added a powerful queue management agent for workflow automation through conversational AI.

New Patterns Added

Queue Management Agent

Queue management assistant using AI agents. Generate structured queues with messages, todos, and attachments through conversation.

We've expanded our Human-in-the-Loop Patterns and artifact capabilities with six new Patterns for interactive AI workflows, data analysis, and web research.

New Patterns Added

HIL Tool Approval Basic

Chat assistant with human approval workflows for safe tool execution.

Table Editor Artifact

Spreadsheet editor with AI chat for data analysis and manipulation. Edit tables through conversation.

HIL Agentic Context Builder

Build brand context through AI-guided questioning and web research. Gathers insights to create detailed brand profiles.

Chart Generation Artifact

Generate burn rate charts and financial analysis through AI chat. Creates interactive visualizations from conversation.

Exa Web Search Tool

Search with Exa's semantic API and real-time crawling. Includes live content extraction and relevance scoring.

Firecrawl Web Search Tool

Crawl and extract content using Firecrawl's API. Includes URL processing, markdown conversion, and structured data extraction.

Firecrawl Scrape Tool Chat

Crawl and extract content using Firecrawl's API. Includes URL processing, markdown conversion, and structured data extraction.

We've introduced a basic workflow pattern to demonstrate AI-powered URL analysis through prompt chaining.

New Patterns Added

Basic Workflow Pattern

Analyze URLs through AI prompt chaining. Scrapes content, analyzes it, and generates summaries.

Workflow - URL Analysis

Analyze URLs through AI prompt chaining. Scrapes content, analyzes it, and generates summaries.

Three new Patterns for web search, audio generation, and PDF analysis expand our AI capabilities.

New Patterns Added

Web Search Tool

Search the web using OpenAI's tool with real-time results. Includes source verification, citation tracking, and contextual answers.

Audio Generation Tool

Convert text to speech using OpenAI's API. Includes multiple voices, speed control, and real-time playback for voiceovers and accessibility.

PDF Analysis Tool

Process PDFs with AI analysis. Extract text, generate summaries, and identify key insights.

We've integrated Google's Gemini 2.5 Flash for advanced text generation, image creation, and image editing capabilities.

New Patterns Added

AI SDK Gemini Flash Text

Generate text and analyze market research using Google's Gemini 2.5 Flash. Includes interactive charts and data visualization.

AI SDK Gemini Flash Image

Generate images using Google's Gemini 2.5 Flash with prompt optimization. Includes style adaptation and customizable parameters.

AI SDK Gemini Flash Image Edit

Generate and edit images using Google's Gemini 2.5 Flash. Create new images from text or edit existing ones with natural language. Includes version history and image comparison.

Enhanced web scraping capabilities with JavaScript rendering and proxy rotation.

New Patterns Added

Jina AI Web Scraper

Scrape websites using Jina AI with JavaScript rendering and proxy rotation. Handles rate limiting and bot detection.

Simple yet powerful web scraping with content extraction and link discovery.

New Patterns Added

Cheerio Web Scraper

Extract content from websites using Cheerio. Includes parsing, link discovery, and rate limiting.

Four new demonstration Patterns showcase tool execution, reasoning, citations, and image generation.

New Patterns Added

Tool Execution Demo

Connect AI agents with external APIs and services. Demonstrates tool calling with weather data, error handling, and visual feedback.

Reasoning Chat Interface

Chat that shows AI's step-by-step reasoning. Includes decision trees, logical breakdowns, and explanatory tooltips.

Inline Citations Demo

Generate content with embedded citations and clickable references. Includes source verification and formatting for research writing.

Image Generation Demo

Generate images using DALL-E 3 with prompt optimization and style presets. Includes real-time preview, downloads, and batch processing.

Four sophisticated agent patterns for parallel processing, multi-step workflows, orchestration, and quality optimization.

New Patterns Added

Parallel Processing Pattern

Process content with multiple AI agents running simultaneously. Demonstrates concurrent analysis for faster results.

Multi-Step Tool Pattern

Execute multi-step workflows with typed tools. Includes streaming, tool chaining, and decision-making for automated tasks.

Orchestrator-Worker Pattern

Coordinate multiple worker agents for project management. Handles task distribution, progress tracking, and result synthesis.

Evaluator-Optimizer Pattern

Improve output quality with dual agents. Evaluator analyzes quality while optimizer refines results through iterative feedback.

Interactive task management with AI workflow automation.

New Patterns Added

Task Management Demo

Task management with AI workflow automation. Creates tasks, tracks progress, and assigns work automatically.

Two foundational chat interfaces for basic conversations and source-cited responses.

New Patterns Added

Basic Chat Interface

Chat interface with streaming responses and message history. Includes typing indicators and message status tracking.

Chat with Citations

Chat with source citations using Perplexity's Sonar model. Answers include clickable links for fact-checking.

Three Patterns for agent routing, prompt engineering, and profile analysis.

New Patterns Added

Agent Routing Pattern

Route user queries to specialized AI agents based on context and intent. Includes dynamic agent selection, load balancing, and fallback handling.

AI SDK Prompt Few-Shot Editor

Create and test prompts with few-shot examples. Evaluate AI responses and tune model performance through experimentation.

AI Profile Enrichment Form

Analyze user profiles with AI and provide personalized suggestions. Enriches data, suggests improvements, and offers career insights.

Welcome to AI SDK Tools - the comprehensive platform for building production-ready AI applications with advanced agent patterns, streaming interfaces, and enterprise-grade components.

What We've Built

AI SDK Tools extends the Vercel AI SDK with enhanced capabilities for building sophisticated AI applications. Our platform provides:

• Production-ready AI agents with orchestrator patterns
• Advanced streaming interfaces for real-time user experiences
• Enterprise-grade components built on shadcn/ui foundation
• Comprehensive Pattern library covering a broad range of AI use cases

🎯 AI Agent Patterns

Orchestrator-Worker Pattern

Complex project coordination through specialized workers with automated task assignment, real-time progress tracking, and project lifecycle management.

Multi-Step Tool Pattern

AI agents that execute complex workflows with real-time web search, iterative workflows, and streaming support for research and analysis automation.

Human-in-the-Loop Pattern

Safe AI tool execution with human oversight, interactive chat interfaces, and tool categorization for critical decision workflows.

Evaluator-Optimizer Pattern

Quality assurance through systematic evaluation with automated content assessment, iterative improvement, and self-correction capabilities.

📊 Marketing & Comparison Tools

Model Comparison

Compare AI models like GPT-4, Claude, and Gemini. Includes filtering, sorting, performance metrics, and pricing data.

Model Comparison Compact

Compare AI models in a compact table format. Includes horizontal scrolling, sortable columns, and condensed data for quick evaluation.

Model Comparison Hover

Compare AI models with hover details for GPT-4, Claude, and Gemini. Includes filtering, sorting, and performance metrics.

📦 Available Patterns

Free Patterns (8)

Marketing & Demos

• AI Model Comparison Table - Comprehensive comparison of leading AI models with performance metrics and pricing analysis
• Integrations Showcase - Dynamic showcase of third-party services with connection capabilities
• Agent ROI Calculator - Calculate return on investment for AI agent implementations
• Changelog Interface - Professional changelog with version history and feature highlights

AI Chat Demos

• ChatGPT Interface - Advanced conversational AI with context management and streaming
• Claude Interface - Sophisticated AI chat with advanced reasoning capabilities
• Grok Interface - Dynamic AI chat with real-time information access

Dashboard & Components

• Dashboard Layout - Complete dashboard with sidebar, charts, and data tables
• New Components Showcase - Latest component demonstrations

Pro Patterns (25+)

AI Agent Patterns

• AI Agents Routing - Intelligent routing system for specialized AI agents
• Human-in-the-Loop - Chat assistant with human approval workflows
• Parallel Processing - Deploy multiple AI agents for concurrent analysis
• Multi-Step Tool Pattern - Complex workflow execution with typed tools
• Orchestrator Pattern - Coordinate multiple worker agents for project management
• Evaluator-Optimizer Pattern - Dual-agent system for quality improvement

Basics - Text Generation

• Generate Text - Simple AI prompt interface with text generation capabilities
• Stream Text - Streaming text generation with real-time response rendering
• Generate Object - Structured object generation with Zod schema validation
• Stream Object - Real-time structured data generation with streaming support

AI Chat Interfaces

• AI Elements Chat - Modern chat interface with streaming and conversation management
• Sources Chat - AI chat with source citations using Perplexity's Sonar model
• Reasoning Chat - Transparent AI chat exposing internal reasoning process
• Task Demo - Interactive task management with AI-powered workflow automation

AI Artifacts & Analysis

• Artifact Table - AI-powered spreadsheet editor with chat interface
• Artifact Chart - Interactive burn rate charts and financial analysis
• Marketing Context Builder - AI-powered marketing strategist with brand context
• PDF Ingest - Intelligent PDF processing with AI-powered document analysis
• Web Search - Intelligent web search with real-time information retrieval
• Profile Enrichment - AI-powered form analysis with personalized suggestions

AI Generation Tools

• Image Generation - DALL-E 3 powered image creation with batch processing
• Audio Generation - Text-to-speech with multiple voice options and emotion modulation
• Gemini Flash Text - Advanced text generation with market research analysis
• Gemini Flash Image - AI image generation with intelligent prompt processing
• Gemini Flash Image Edit - Generate and edit images with natural language

Data & Scraping Tools

• Cheerio Scraper - Robust web scraping with intelligent parsing
• Jina Scraper - Advanced scraping with anti-detection capabilities
• Inline Citation - Factually accurate content with embedded source citations

Advanced Tools

• Prompt Few-Shot - Prompt engineering tool for few-shot learning
• Tool Integration - Comprehensive platform for connecting AI agents with external APIs
• Web Search Tools - Multiple search providers (Claude, Exa, Firecrawl)

Marketing & Comparison

• Model Comparison 2 - Interactive comparison tool for leading AI models
• Model Comparison Compact - Compact table-based model comparison
• Model Comparison Hover - Interactive comparison with hover interactions

🛠️ Technical Foundation

Built on Proven Stack

• Vercel AI SDK - Core AI functionality and streaming
• shadcn/ui - Production-ready component library
• TypeScript - Type-safe development experience
• Tailwind CSS - Utility-first styling

Advanced Features

• Streaming Output - Real-time text and artifact streaming
• Tool Calling - Structured AI tool execution
• State Management - Global state with Zustand integration
• Development Tools - Comprehensive debugging capabilities

Enterprise Ready

• Authentication - Secure access control for Pro Patterns
• API Integration - Connect with external services
• Error Handling - Robust error recovery and fallback mechanisms
• Performance - Optimized for production workloads

🎉 What's Next

Our roadmap includes:

• More AI Agent Patterns - Additional orchestration and workflow patterns
• Enhanced Artifacts - Advanced visual components and data visualization
• Integration Expansions - More third-party service connections
• Performance Optimizations - Faster streaming and reduced latency
• Developer Experience - Better tooling and debugging capabilities

🚀 Get Started

Ready to build the future of AI applications?

• Free Patterns are available immediately
• Pro Patterns require authentication for access
• All Patterns include complete source code and documentation
• Copy and paste directly into your projects

Start building with AI SDK Tools today and ship AI features 10x faster.

Enhanced web search capabilities with Claude AI integration.

New Patterns Added

Claude Web Search Tool

Search the web with Claude AI. Includes result filtering, multiple search types, and language support.

_Built with ❤️ by the nolansym