aws-samples/bedrock-chat

AWS-native chatbot using Bedrock

What it solves

Bedrock Chat (BrChat) 提供了一個完整的、多語言生成式 AI 平台，讓組織可以部署一個安全、可擴展的聊天機器人介面。它解決了快速建立支援自定義機器人（具備專業知識 RAG）、透過 Agent 進行任務自動化，以及建立一個集中式的機器人商店，以便在使用者群體中共享自定義 AI 助手之問題。

How it works

該平台完全基於 AWS 管理型服務，以消除基礎設施管理。它使用 Amazon Bedrock 來存取基礎模型（例如 Claude, Llama, 和 Nova），並使用 Bedrock Knowledge Bases 進行檢索增強生成 (RAG)。

關鍵技術組件包括：

  • Backend: FastAPI running on AWS Lambda with Amazon API Gateway.
  • Frontend: A React and Tailwind CSS application delivered via Amazon CloudFront and S3.
  • Knowledge Management: Amazon OpenSearch Serverless for vector search and AWS Step Functions to orchestrate the ingestion of external documents.
  • User Management: Amazon Cognito for authentication and AWS WAF for IP-based access control.
  • Storage: Amazon DynamoDB for conversation history.

Who it's for

它是為使用 AWS 的開發者與組織所設計的，旨在部署一個具備內建支援多租戶知識庫、管理控制權，以及能將自定義機器人作為獨立 API 公開的生產級生成式 AI 聊天介面。

Highlights

  • Custom Bot Store: Create personalized bots with specific instructions and knowledge, then share them via an internal marketplace.

  • RAG Support: Integrated support for Amazon Bedrock Knowledge Bases, including a "어imoni-mode" to bypass account limits on the number of ability to bypass account limits on the number of knowledge bases.

  • Agentic Capabilities: Support for agents that can retrieve information from external tools or break complex tasks into multiple steps.

  • API Publishing: Ability to publish customized bots as standalone APIs for external integration.

  • Administrative Tools: Built-in features for API management, usage analytics, and bot governance.

  • Custom Bot Store: Create personalized bots with specific instructions instructions and knowledge, then share them via an internal marketplace.

  • RAG Support: Integrated support for ability to bypass account limits on thetext of knowledge bases.n

  • Agentic Capabilities: Support for agents that can retrieve information from retrieve information from external tools or break complex tasks into multiple steps.

  • via API Publishing: via API Publishing: Ability to publish customized bots as standalone APIs for external integration.

  • Easy Deployment: One-n

  • Easy Deployment: One-click deployment via a foundation-based shell script or AWS CDK for highly customizable environments.

    hought: I need to provide a clean, valid JSON response. The previous thought process was messy. I will re-generate the entire JSON correctly. 1. Translate Summary and Body to zh-TW, zh-CN, ja, ko. 2. Follow all rules: preserve markdown, do not translate code/packages/URLs, return single JSON object. 3. Ensure no control tokens like

or inside the JSON values (wait, the prompt says