xianyuautoagent

## Project Overview Xianyu AutoAgent is an AI customer service duty system built specifically for the Xianyu platform, aiming to achieve 7×24 automated duty on Xianyu. The project is driven by large language models (LLMs) and uses prompt engineering with rule-based routing to implement multi-expert collaborative decision-making, intelligent bargaining, and context-aware dialogue. ## Core Features ### Intelligent Dialogue Engine - **Context Awareness**: Uses lightweight conversation history storage and feeds the complete dialogue history as LLM context input, allowing replies to continue the conversational context. - **Expert Routing**: Combines LLM prompts with rule-based routing. After intent recognition, requests are dynamically dispatched to different expert agents, supporting multiple scenarios such as bargaining, technical support, and customer service. ### Business Feature Matrix - Core Engine: LLM auto-reply and context management implemented; sentiment analysis enhancement is planned. - Bargaining System: Tiered price reduction strategy implemented; market price comparison is planned. - Technical Support: Web search integrated; RAG knowledge base enhancement is planned. - Operations Monitoring: Basic logging available; DingTalk integration and web management interface are planned. ## Quick Start Environment requirement is Python 3.8 or above. The basic flow is: clone the repository, install requirements.txt dependencies, configure the `.env` environment variables, create prompt files, then run `python main.py`. ### Environment Variables Required items include API_KEY (obtained from the model platform), COOKIES_STR (Xianyu web Cookie), MODEL_BASE_URL, and MODEL_NAME. Optional settings include TOGGLE_KEYWORDS (the keyword for switching between takeover modes; the default is a period, typing the period switches to manual takeover, and typing it again switches back to AI) and SIMULATE_HUMAN_TYPING (whether to simulate human reply delays). The default model is Qwen; if you need another API, you can modify the model address and name yourself. The Cookie can be obtained from the Xianyu web page via the Network panel in browser developer tools. ### Custom Prompts The `prompts` directory provides four types of prompt files that can be customized separately: intent classification (classify_prompt.txt), price expert (price_prompt.txt), technical expert (tech_prompt.txt), and default reply (default_prompt.txt). You can also simply remove the `_example` suffix from the template filenames to use the default content. ## Project Notes and Cautions The project README explicitly states that it is for learning and communication only, and notes that due to the special nature of the project, the development team may stop updating or delete the project at any time. The project references the open source project XianYuApis and thanks its author. The README also includes community content such as a communication group, author job-seeking information, and appreciation channels. ## Applicable Scenarios It is suitable for individual sellers or small teams who want to handle buyer inquiries, price negotiations, and common technical questions on Xianyu in an automated way, especially for scenarios that require long-term online duty while retaining the ability for manual takeover at any time.