> browse/registry

> structured-output

Force LLMs to return typed, validated JSON — not free-text. Use when someone asks to "get structured data from LLM", "parse LLM response as JSON", "make AI return typed output", "validate LLM output", "extract structured data with AI", "use instructor with OpenAI", or "get reliable JSON from Claude/GPT". Covers OpenAI structured outputs, Anthropic tool_use for structured data, Instructor library, Zod schemas, Pydantic models, and retry strategies for malformed responses.

> llm-cost-optimizer

Track, analyze, and reduce LLM API costs — model routing, prompt caching, semantic caching, and budget alerts. Use when someone asks to "reduce AI costs", "track LLM spending", "optimize API costs", "set up model routing", "cache LLM responses", "compare model costs", "set budget limits for AI", or "my OpenAI bill is too high". Covers cost tracking per feature/user, smart model routing (expensive model for hard tasks, cheap for easy), semantic caching, prompt compression, and budget alerting.

> markdown-new

Convert any public URL into clean, LLM-ready Markdown using the markdown.new service. Use for content extraction, RAG ingestion, article summarization, research, archiving, and token-efficient web reading.

> continue-dev

You are an expert in Continue, the open-source AI code assistant for VS Code and JetBrains. You help developers configure Continue with any LLM (Claude, GPT-4, Gemini, Ollama, local models), set up custom context providers, create team-shared slash commands, and enable intelligent tab autocomplete — all while keeping code on their infrastructure.

> openai-agents

You are an expert in the OpenAI Agents SDK (formerly Swarm), the official framework for building multi-agent systems. You help developers create agents with tool calling, guardrails, agent handoffs, streaming, tracing, and MCP integration — building production-grade AI agents that coordinate, delegate tasks, and execute tools with built-in safety controls.

> localai

Expert guidance for LocalAI, the open-source drop-in replacement for OpenAI's API that runs locally. Helps developers self-host LLMs, image generators, audio transcription, and text-to-speech models with an OpenAI-compatible API — no GPU required, completely offline and private.

> deepeval

Expert guidance for DeepEval, the open-source framework for unit testing LLM applications. Helps developers write test cases, define custom metrics, and integrate LLM quality checks into CI/CD pipelines using a pytest-like interface.

> prompt-tester

Design, test, and iterate on AI prompts systematically using structured evaluation criteria. Use when building AI features, optimizing agent instructions, comparing prompt variants, or evaluating output quality across edge cases. Trigger words: prompt engineering, prompt testing, eval, LLM evaluation, prompt comparison, A/B test prompts, prompt optimization, system prompt, instruction tuning.

> scikit-learn

Assists with building, evaluating, and deploying machine learning models using scikit-learn. Use when performing data preprocessing, feature engineering, model selection, hyperparameter tuning, cross-validation, or building pipelines for classification, regression, and clustering tasks. Trigger words: sklearn, scikit-learn, machine learning, classification, regression, pipeline, cross-validation.

> llamaindex

Assists with building RAG pipelines, knowledge assistants, and data-augmented LLM applications using LlamaIndex. Use when ingesting documents, configuring retrieval strategies, building query engines, or creating multi-step agents. Trigger words: llamaindex, rag, retrieval augmented generation, vector index, query engine, document loader, knowledge base.

> lm-studio-subagents

Offload tasks to local LLMs via LM Studio. Use when a user asks to run local models with LM Studio, save API costs by using local LLMs, create subagents with local models, offload summarization or classification to a local model, or use LM Studio's API for batch processing. Covers local model inference, task delegation, and cost optimization.

> mcp-sdk

You are an expert in MCP (Model Context Protocol), the open standard by Anthropic for connecting AI models to external tools and data sources. You help developers build MCP servers that expose tools, resources, and prompts to any MCP-compatible client (Claude Desktop, Cursor, Windsurf, Cline, Continue) — creating a universal plugin system for AI assistants.

> anthropic-sdk

Integrate Claude AI into applications with the Anthropic SDK. Use when a user asks to add Claude to an app, use Claude for text generation, build a chatbot with Claude, use Claude's long context window, implement tool use with Claude, stream Claude responses, use Claude for code generation, document analysis, or reasoning tasks. Covers Messages API, streaming, tool use, vision, system prompts, extended thinking, and batch processing.

> comet-ml

Expert guidance for Comet ML, the platform for tracking machine learning experiments, managing models, and monitoring production ML systems. Helps developers log experiments, compare model versions, and build reproducible ML pipelines with automatic code/data versioning.

> gcp-bigquery

Analyze massive datasets with Google BigQuery. Run SQL queries on petabytes of data, load and stream data in real-time, create materialized views, and use BigQuery ML for machine learning models directly in SQL.

> huggingface

Work with Hugging Face's ecosystem for machine learning — transformers library, model hub, tokenizers, inference pipelines, and fine-tuning. Covers downloading pre-trained models, running inference, training custom models, and publishing to the Hub.

> guidance

You are an expert in Guidance, Microsoft's library for controlling LLM output with constrained generation. You help developers write programs that interleave text generation with control flow (loops, conditionals, regex constraints, JSON schemas, function calls) — ensuring LLM output always matches the expected format by constraining the token generation process itself, not just prompting.

> instructor

You are an expert in Instructor, the library for getting structured, validated output from LLMs. You help developers extract typed data from unstructured text using Pydantic models (Python) or Zod schemas (TypeScript), with automatic retries on validation failures, streaming partial objects, and support for OpenAI, Anthropic, Google, and local models — turning LLMs into reliable data extraction engines.

> job-queue

Build job queues and background worker systems using BullMQ, Celery, or Sidekiq. Use when you need to offload slow tasks from request handlers — email sending, PDF generation, image processing, data exports, or any work that takes more than a few hundred milliseconds. Covers job priorities, concurrency control, scheduled jobs, progress tracking, and graceful shutdown. Trigger words: background job, worker, queue, async task, BullMQ, Celery, cron job, scheduled task, job retry.

> ollama-sdk

You are an expert in Ollama, the tool for running open-source LLMs locally. You help developers run Llama, Mistral, Gemma, Phi, CodeLlama, and other models on their machine with a simple CLI and REST API — enabling private AI development, offline inference, fine-tuning experiments, and cost-free prototyping without sending data to cloud APIs.