cascade-sdk 0.2.0b14

Agent observability SDK for tracking AI agent execution

Cascade SDK

Agent observability platform for tracking AI agent execution, LLM calls, and tool usage.

Quick Start

Install from PyPI

pip install cascade-sdk

Setup

1. Set your API key (get one from Cascade Dashboard):

   export CASCADE_API_KEY="your-api-key"
   

2. Use in your code:

   from cascade import init_tracing, trace_run, wrap_llm_client, tool
   from anthropic import Anthropic
   import os
   
   # Initialize tracing (uses cloud endpoint by default)
   init_tracing(project="my_project")
   
   # Wrap LLM client
   client = wrap_llm_client(Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY")))
   
   # Decorate tools
   @tool
   def my_tool(query: str, client) -> str:
       """My custom tool."""
       response = client.messages.create(
           model="claude-3-haiku-20240307",
           max_tokens=100,
           messages=[{"role": "user", "content": query}]
       )
       return response.content[0].text
   
   # Trace agent execution
   with trace_run("MyAgent", metadata={"task": "example"}):
       result = my_tool("What is AI?", client)
       print(result)
   

3. View traces in the Cascade Dashboard

TypeScript (Node.js)

This repo also includes a Node.js TypeScript SDK under typescript/cascade-sdk/.

• Docs: docs/INTEGRATIONS_TYPESCRIPT_SDK.md
• Vercel AI SDK (TypeScript) OTEL export: docs/INTEGRATIONS_VERCEL_AI_SDK.md

Features

• ✅ Zero setup - No backend services to run
• ✅ Cloud-first - Traces automatically sent to cloud
• ✅ LLM tracking - Automatic tracking of LLM calls (Anthropic, OpenAI, etc.)
• ✅ Tool tracing - Decorate functions with @tool for automatic tracing
• ✅ Rich metadata - Add custom metadata to traces
• ✅ OpenTelemetry - Built on OpenTelemetry standards
• ✅ Vercel AI SDK support - Ingest TypeScript OTEL spans (see docs/INTEGRATIONS_VERCEL_AI_SDK.md)

Configuration

Environment Variables

• CASCADE_API_KEY - Your Cascade API key (required)
• CASCADE_ENDPOINT - Override default endpoint (default: https://api.runcascade.com/v1/traces)

Custom Endpoint

If you need to use a custom endpoint:

init_tracing(
    project="my_project",
    endpoint="https://your-custom-endpoint.com/v1/traces",
    api_key="your-api-key"
)

CLI Commands

• cascade info - Show setup instructions and information
• cascade --help - Show help message
• cascade --version - Show version

Development

This is a monorepo containing:

• cascade/ - SDK package (published to PyPI)
• backend/ - Backend service (not included in package)
• dashboard/ - Frontend dashboard (not included in package)

Install SDK in Development Mode

pip install -e .