Agen & Multi-Agen

Claude Quickstarts: support, browser and agent reference applications

anthropics/claude-quickstarts

Anthropic’s reference application collection covers knowledge-backed support, data analysis, browser use and desktop interaction as starting points for development.

★ 17,7KBintang
⑂ 3KFork
197Isu terbuka
TypeScriptBahasa
MITLisensi
Q84Skor editorial

Ringkasan

Claude Quickstarts contains independent application directories, each with its own setup and service requirements. The examples go beyond a single model call to show interfaces, tool execution, knowledge retrieval and task state. Select the directory closest to your use case, understand how inputs, tools and responses flow through it, then replace the sample data and application rules.

Fitur utama

  • Includes a knowledge-backed customer support interface.
  • Includes an interactive data analysis application.
  • Provides a containerized computer-use demo.
  • The browser demo uses a Playwright-backed tool.
  • Includes a coding-agent example with progress across sessions.
  • Each application directory provides its own README and setup guidance.

Persyaratan, instalasi, dan mulai cepat

1. Clone anthropics/claude-quickstarts and select an application directory.
2. For customer-support-agent, enter that directory and run npm install.
3. Configure ANTHROPIC_API_KEY and the required AWS credentials in .env.local as documented.
4. Create an Amazon Bedrock knowledge base and update the sample’s knowledge-base identifiers and names.
5. Check that the example model identifiers remain available to your account and update them as needed.
6. Run npm run dev and test retrieval and answers with questions whose answers are known.

Penggunaan

For a support assistant, start with one product manual and test questions about returns or setup. Verify that answers point to the right source passages, and record missing information or cases requiring human support. For browser automation, switch to browser-use-demo and follow its own environment and tool setup; the support app’s commands are not a universal installation recipe for the collection.

Implementation notes
These are development starting points. A deployed service still needs authentication, access controls, error handling and usage monitoring. Review sample model names, regions and knowledge-base identifiers against your own account.

Kompatibilitas model dan kasus penggunaan

Requirements vary by subproject. The support app needs Node.js, Claude API access and an AWS Bedrock knowledge base. Desktop and browser demos have additional dependencies. Use an isolated development environment and server-side credential handling.

Catatan lisensi dan risiko

Example code uses MIT. Model APIs and external services such as AWS have separate billing and terms.

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython

RAGFlow

infiniflow/ragflow

★ 86,7KGo