pocketgroq

Wraps the Groq API client to provide streaming and non-streaming text generation with configurable model selection, temperature, and token limits. Abstracts authentication and request formatting, allowing developers to call Groq's inference endpoints without managing raw HTTP or SDK boilerplate. Supports both synchronous completion calls and streaming responses for real-time token output.

Implements structured chain-of-thought prompting by decomposing complex queries into intermediate reasoning steps before final answer generation. Uses prompt templates that explicitly request step-by-step thinking, then chains multiple API calls together where each step's output feeds into the next. Enables more accurate problem-solving for mathematical, logical, and multi-step reasoning tasks by forcing the model to show its work.

Combines web scraping (likely using BeautifulSoup or similar) with Groq API calls to extract and summarize relevant information from web pages. Fetches raw HTML, parses it, and uses the LLM to identify and extract structured data or summaries from unstructured web content. Enables semantic understanding of web pages without manual parsing rules.

Integrates web search (likely Google Search API or similar) with Groq text generation to retrieve current information and synthesize it into coherent answers. Performs a search query, retrieves top results, and uses the LLM to summarize or synthesize findings into a single response. Enables agents to access real-time information beyond their training data cutoff.

Provides a framework for building autonomous agents that can call tools (web search, scraping, code execution, etc.) in a loop until a goal is reached. Uses the LLM to decide which tool to call next based on current state, executes the tool, and feeds results back to the LLM for next-step planning. Implements a reasoning loop where the agent iteratively refines its approach based on tool outputs.

Provides a templating system for constructing dynamic prompts with variable substitution, allowing developers to define reusable prompt patterns with placeholders for context, user input, or system state. Supports string formatting or template engines to inject values at runtime, enabling consistent prompt structure across multiple queries without string concatenation.

Handles Groq API errors, timeouts, and malformed responses with structured error messages and fallback behavior. Parses JSON responses from the API, validates structure, and provides meaningful error context when parsing fails. Abstracts away raw HTTP error codes and API-specific error formats into developer-friendly exceptions.

Maintains conversation history across multiple turns, managing context window constraints by truncating or summarizing older messages when the conversation exceeds token limits. Implements sliding window or summarization strategies to keep recent context while staying within Groq's token limits. Enables multi-turn conversations without losing context or exceeding API constraints.