Skill UI

This skill details advanced techniques for enhancing the effective context capacity of LLMs. It outlines four core strategies—Compaction, Observation Masking, KV-Cache Optimization, and Context Partitioning—to prevent context limits from constraining task complexity. By strategically managing token usage, users can reduce operational costs, minimize latency, and implement robust, long-running agent systems for handling large documents or conversations.

This skill manages and refines search results obtained from Exa's neural search API. It provides comprehensive control over content extraction scope, allowing users to selectively retrieve metadata, highlights, full text, or structured summaries. Key features include implementing result caching with TTL, deduplicating citations, and crucial token budget management, ensuring the retrieved context fits within LLM context window limits for robust RAG pipelines.

This skill provides comprehensive methods to significantly reduce expenditure when using the Ideogram API for image generation. Techniques include implementing two-phase generation workflows (drafting with cheap models, finalizing with premium models), batching multiple images in a single API call, caching identical prompts, and setting up robust budget tracking alerts. Ideal for developers managing large-scale creative AI projects.

An expert guide and framework for building production-grade, sophisticated AI agent systems. This skill specializes in utilizing the latest LangChain 0.1+ and LangGraph APIs to implement multi-agent orchestration, advanced Retrieval-Augmented Generation (RAG) pipelines (e.g., RAG Fusion, HyDE), and complex state management. It covers comprehensive topics from architectural design (Plan-and-Execute, ReAct Agents) to production deployment (FastAPI streaming, LangSmith observability, caching, and robust error handling).

Mamba is a selective state-space model architecture delivering O(n) inference for million-token sequences, offering hardware-aware Mamba-1 and Mamba-2 variants that beat Transformers on speed without KV cache overhead.

This guide provides advanced techniques for optimizing Mistral AI API usage, focusing on latency reduction and throughput maximization. Learn to implement robust caching strategies, utilize streaming responses (SSE), select the optimal model for specific use cases (e.g., Mistral Small for chat, Codestral for code), manage concurrent requests using rate limiting, and efficiently use the Batch API. Essential for building high-performance, production-grade AI applications.

This skill provides advanced strategies for managing large language model (LLM) context windows and token usage via OpenRouter. Functions include context window lookup, context-aware model selection, conversation trimming, document chunking for RAG, and prompt caching. Use this when dealing with long conversations or large documents to maximize quality while minimizing API costs and preventing context limit overflow.

A comprehensive toolkit designed to manage the complex data flow associated with Perplexity search workflows. This skill implements critical security measures, including PII sanitization in user queries, validates external citations for accuracy, optimizes performance via result caching with freshness policies, and manages conversation context to maintain coherence and reliability. Essential for building robust, compliance-aware AI applications.

This guide provides comprehensive instructions for deploying applications powered by the Perplexity Sonar API. It covers setup for major cloud platforms like Vercel, Google Cloud Run, and Docker. Users can deploy production-ready endpoints, manage required secrets, implement edge functions, and utilize caching mechanisms (e.g., Redis) to ensure high availability and performance in production environments.

This guide outlines the comprehensive reference architecture for building advanced, AI-powered search and research applications, modeled after Perplexity. It covers three core tiers: a searchable service with model routing and caching, a citation pipeline for validation and rendering, and an orchestrator for multi-query research synthesis. Use this when designing complex, production-grade, search-augmented LLM integrations or reviewing advanced system patterns.

Covers prompt and response caching plus Cache Augmented Generation patterns for LLMs such as Claude and OpenAI to reuse stable context, reduce repeat work, and cut latency/cost on recurring queries.

SGLang is a high-performance serving framework for LLMs and VLMs that uses RadixAttention prefix caching to deliver structured outputs (JSON/regex/grammar) and agentic workflows with function calls, providing up to 5× faster inference than vLLM across multi-GPU production deployments.