agentwatchx 0.2.2

Observability SDK for AI Agent Execution — catches hallucinations, silent failures, and missing executions

AgentWatchX Python SDK

Observability for AI agents. Catches hallucinations, silent failures, and missing executions automatically.

Install

pip install agentwatchx

2-Line Integration

import agentwatchx
agentwatchx.init(api_key="your_api_key")

# That's it. Every OpenAI/Anthropic/Gemini/Groq/Mistral/LangChain/LlamaIndex call is now traced.
# No decorators, no wrappers, no code changes needed.

What it captures

• Model, input, output, token usage, latency — automatically
• Tool calls made by the LLM (function calling)
• Hallucination detection — agent claims actions it never performed
• Failure masking — tool errors the agent hides from users
• Missing execution — user asks for action, agent just talks

Supported Libraries

Library	Auto-instrumented
OpenAI	✅
Anthropic	✅
Gemini	✅
Groq	✅
Mistral	✅
LangChain	✅
LlamaIndex	✅

Configuration

agentwatchx.init(
    api_key="your_api_key",
)

Advanced Usage

# Manual trace decorator
@agentwatchx.trace
def my_agent(query: str):
    return client.chat.completions.create(...)

# Manual logging
agentwatchx.log({"service": "my-agent", "input": "hello", "output": "world"})

# Check status
print(agentwatchx.status())
# → {"initialized": True, "instrumented": ["openai"], "buffered_traces": 0}

Links

• Documentation
• Dashboard