ai-wingman

The right* way to wirestreaming chatinto your Next.js app.

Scaffolds production-ready AI patterns into your existing project — chat, RAG, agents, tools, audio, multimodal, evals, and more. No lock-in. Just plain TypeScript files you own.

npx ai-wingman add chat

How it works

ai-wingman wires together Vercel AI SDK, shadcn/ui, and ai-elements into 18 battle-tested AI patterns — then generates the files and exits. There is no runtime dependency on ai-wingman after scaffolding. Pick the pattern that matches your use case, or combine several to build a complete product.

Use with your AI coding assistant

Add the ai-wingman skill file to your AI coding assistant so it knows how to scaffold patterns for you — which command to run, which flags to pass, and how to combine patterns for common use cases.

npx ai-wingman skill

Downloads SKILL.md to your current directory. Add it to your assistant's context or skills directory — once loaded, your assistant will use wingman whenever you ask it to add an AI feature to your project.

Patterns

chat
Multi-turn streaming chat with conversation history.
streamingconversationstorage
{}
structured-output
Extract a typed, validated JSON object from any prompt.
typed JSONzodhook
tools
Let the model call your functions, APIs, and databases.
tool callingAPI callsside effects
{:::}
stream-object
Structured JSON that renders field-by-field as it streams.
streamingtyped JSONprogressive
multimodal
Chat with image and file attachments, analysed by vision models.
imagesvisionfile upload
audio
Speech-to-text input, text-to-speech output, or both.
voiceSTTTTS
document-processing
Upload a PDF or file, extract typed structured data — wired end to end.
PDFfile uploadextraction
generative-ui
Tool calls render as React components — UI builds progressively as the model generates it.
streamingtoolscomponents
image-gen
Prompt-to-image generation with display and download.
text-to-imagestreaming

Use cases

Most real products combine two or more patterns. Here are ten reference use cases — each one maps directly to a set of patterns you can scaffold with a few commands.

Customer support copilot

Answers questions from your docs, remembers each user's history across sessions, and blocks policy-violating messages before they reach the model.

Voice-powered knowledge base

Speak a question, get an answer grounded in your documents. Transcribe audio input, retrieve the most relevant chunks, stream the grounded response.

Contract review pipeline

Upload a PDF, extract key terms and clauses as typed JSON, flag risky content against a policy, and require explicit human sign-off before proceeding.

Personalised research assistant

An autonomous agent that runs multi-step research, surfaces the most relevant results via hybrid keyword + vector search, and remembers your preferences across sessions.

Async research report

Kick off a long-running research task without hitting serverless timeouts. An orchestrator delegates sub-topics to specialist agents and synthesises the findings.

Generative dashboard

A natural language query triggers tool calls that fetch live data. Results stream back and render progressively as cards, charts, and tables — no page reload.

Photo-to-structured data

Upload a photo of a receipt, business card, or whiteboard. The model reads the visual content and returns a typed, validated JSON object your app can use directly.

Content creation studio

Draft copy and generate matching visuals from a single prompt. Stream the text field-by-field as it generates while the image renders alongside it.

Adaptive onboarding flow

An assistant that remembers what each user has already completed and renders personalised next-step components — checklists, tips, or prompts — as the conversation progresses.

AI deployment quality gate

Every pull request runs a dataset eval. A structured output captures per-record judge scores, pass/fail status, and mean score — ready to log, diff, or display in a dashboard.

Is this right for you?

Good fit

  • Streaming chat assistant or copilot
  • Structured data extraction from input or documents
  • PDF and file document intelligence pipeline
  • Agent that calls your APIs from natural language
  • Knowledge base Q&A grounded in your own data
  • Multimodal features that understand images and files
  • Voice interfaces — speech-to-text, text-to-speech
  • Automated quality gates for AI outputs
  • Per-user long-term memory across sessions
  • Hybrid keyword + vector search with reranking
  • Content moderation and policy classification layer
  • AI-generated UI components streamed in real time
  • Image generation with display and export
  • Multi-agent system with orchestrator + specialist workers
  • Background agent running outside the request cycle
  • Video generation with prompt and playbackcoming soon
  • AI layer for real-time collaborative editingcoming soon
  • React Native and Expo mobile appscoming soon
  • Vue, Svelte, and Angular supportcoming soon

Not a fit

  • Model training, fine-tuning, or data science pipelines
    A completely different toolchain (Python, PyTorch, Jupyter) that is already well-served by dedicated frameworks.