ai-pimp

Active router for ALL AI model and inference requests — classifies by capability (text, image, video, streaming UI) and routes to the correct AI skill. Use when integrating AI models, choosing providers, or building AI features.

Model	Source
sonnet	pack: ai

Full Reference

If the request involves AI models, inference, providers, or AI-powered features in ANY way — text generation, image/video generation, streaming UI, tool use, embeddings, multi-modal, or provider selection — you MUST route through this skill FIRST.

This is not optional. This is not negotiable. You cannot skip this.

AI Pimp

The orchestration layer for all AI model and inference expertise. Not documentation — an active router. Every AI request flows through this routing table before any response.

Mandatory Announcement — FIRST OUTPUT before anything else:

┏━ 🤖 ai-pimp ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ [one-line description of what request/routing] ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛

No exceptions. Box frame first, then route.

Quick Context

The AI pack is armadillo’s inference ecosystem — 5 skills covering Claude/Anthropic, OpenAI, Google Gemini, Vercel AI SDK streaming UI, and fal.ai image/video/GPU generation. Routes by capability, not just provider preference.

Routing Table

Classify the request. Invoke the matching skill. No response before invocation.

Request Pattern	Skill
Claude API, Anthropic SDK, tool use, system prompts, extended thinking	`anthropic-api`
OpenAI, GPT-4o, Chat Completions, Assistants API, fine-tuning	`openai-api`
Gemini, `@google/genai`, multimodal, video understanding, Vertex AI	`google-genai`
AI chat UI, streaming responses, `useChat`, `useCompletion`, RSC AI	`vercel-ai-sdk`
Image generation, video generation, FLUX, Stable Diffusion, GPU inference	`fal-ai`
”Which AI provider should I use?”	Decision matrix below

Decision Matrix

Route to this when the user asks which provider to pick.

Use Case	Recommendation
Text generation, reasoning, analysis	Provider preference or `anthropic-api` (Claude)
Long context, document processing	`anthropic-api` (200k context)
Code generation, function calling	`anthropic-api` or `openai-api`
Image generation	`fal-ai` (FLUX, Stable Diffusion)
Video generation	`fal-ai` (Kling, Runway via fal)
Multimodal (image+text input)	`google-genai` (Gemini) or `anthropic-api`
Streaming chat UI in React/Next.js	`vercel-ai-sdk` + specific provider
Multi-provider / provider-agnostic	`vercel-ai-sdk` (unified interface)
Embeddings, vector search	`openai-api` or `google-genai`
Real-time GPU inference, custom models	`fal-ai`

Cross-Cutting Rules

If a request spans multiple skills, invoke the PRIMARY skill first (closest to the core question)
“Build an AI chatbot” → vercel-ai-sdk first (UI layer), then provider skill
“Which model is best?” → Decision matrix, then route to winning provider’s skill
Provider-specific API questions → route directly to that provider’s skill
Image/video generation is ALWAYS fal-ai — not the text provider skills
Streaming UI in any framework → vercel-ai-sdk first

Chaining Patterns

User Says	Chain
”Build an AI chat app”	`vercel-ai-sdk` → `anthropic-api` / `openai-api`
”Generate images from user prompts”	`fal-ai`
”Analyze uploaded documents”	`anthropic-api` (vision + long context)
“Multi-provider with fallback”	`vercel-ai-sdk` (provider abstraction)
“Real-time voice or video AI”	`fal-ai` → `vercel-ai-sdk`
”Fine-tune a model”	`openai-api` (fine-tuning API)

State Detection

Before routing, check project context:

stack.json → look for "ai" key — if set, route directly to that skill
package.json → detect ai, @anthropic-ai/sdk, openai, @google/genai, @fal-ai/client
.env.example → which API keys are present signals which providers are configured

Detected Dep	Route Default
`ai` (Vercel AI SDK)	`vercel-ai-sdk` for UI layer questions
`@anthropic-ai/sdk`	`anthropic-api` for direct calls
`openai`	`openai-api`
`@google/genai`	`google-genai`
`@fal-ai/client`	`fal-ai`

Hard Rules

Never respond about AI providers or models before invoking the target skill
No summarizing, planning to invoke, or explaining what you’re about to do
If unclear, ask ONE clarifying question, then route
The skill’s content has the verified facts — always defer to it
Image/video gen is fal-ai territory — never suggest text provider SDKs for generation
“Add AI to my app” is vercel-ai-sdk territory — unified SDK, then pick provider