iamzulx/crypto-rag

Asisten crypto berbahasa Indonesia: RAG pengetahuan 267 topik + data pasar realtime (6 bursa, WebSocket, derivatif, on-chain, TVL, DeFi) + tool-calling agent + LLM synthesis

What it solves

This project provides an Indonesian-language crypto assistant that solves the problem of outdated information in AI assistants. By separating static knowledge (educational concepts) from volatile market data (prices, funding rates, TVL), it ensures that users receive accurate, real-time financial data without the risk of LLM hallucinations regarding prices.

How it works

The system uses a router to detect user intent and directs queries to different data sources based on the type of request:

  • Static Knowledge: Uses a hybrid retrieval system (BM25 and FAISS dense embeddings) with Reciprocal Rank Fusion to pull educational content from a knowledge corpus of over 146 topics.
  • Real-time Market Data: Fetches live data directly from 6 exchanges (Binance, OKX, Bybit, KuCoin, Kraken, Coinbase) and on-chain sources (DefiLlama, Mempool.space) via tool-calls, bypassing the LLM for numerical values.
  • LLM Synthesis: Optionally combines the retrieved knowledge and live data into a natural language response using an OpenAI-compatible endpoint.
  • Agentic Workflow: A tool-calling agent can handle complex, multi-step queries by automatically selecting the necessary tools to answer a request.

Who it’s for

Indonesian-speaking crypto enthusiasts, traders, and learners who need a combination of educational guidance and live market monitoring (including portfolio tracking and price alerts) in a single interface.

Highlights

  • Keyless Market Data: Accesses spot prices, funding rates, and TVL without requiring API keys.
  • Hybrid RAG: Combines keyword search and semantic search for better retrieval precision.
  • Conditional Price Alerts: Supports streaming live prices with alerts for specific price thresholds (e.g., BTC>65000).
  • On-chain Integration: Provides live Bitcoin mempool fees, Ethereum gas fees, and Solana supply data.
  • Portfolio Tracker: Built-in tool to track holdings and calculate live P/L.
  • LLM Agnostic: Compatible with various providers like Groq, Ollama, and LM Studio.

Related

  • Project
    fufankeji/deepseek-harness-studio

    A zero-code desktop application for DeepSeek Harness that provides a visual interface for managing AI agents, plugins, and local model inference.

  • Project
    zhongyu09/openchatbi

    An open-source, chat-based BI tool that uses LLMs to convert natural language into SQL, enabling users to analyze and visualize database data through conversation.

  • Project
    qhkm/zeptoclaw

    A fast, secure, and local-first personal AI assistant infrastructure written in Rust, providing sandboxed autonomy and multi-provider LLM support in a single 6MB binary.

  • Project
    Azure/co-op-translator

    An AI-powered tool that keeps multilingual GitHub documentation current by automating the translation of Markdown, Jupyter notebooks, and images while preserving repository structure and links.