SDK Auto-Discovery

Agents instrumented with the NodeLoom SDK are automatically added to your inventory — no configuration needed.

How It Works

When your SDK sends a trace_start event, NodeLoom extracts the agent name, framework, language, and SDK version. It creates or updates the agent in your inventory automatically.

As spans arrive, NodeLoom also tracks dependencies:

• LLM spans → model name and provider (e.g., gpt-4o / OpenAI)
• Tool spans → tool name (e.g., knowledge-base-search)

Framework Detection

The SDK automatically detects which AI framework is installed in your environment and includes it in every trace. Supported frameworks:

• LangChain — Python, TypeScript
• CrewAI — Python
• AutoGen — Python
• LlamaIndex — Python
• LangChain4j — Java
• Custom — any framework not in the list above

Example

A standard SDK trace automatically populates the inventory:

from nodeloom import NodeLoom, SpanType

client = NodeLoom(api_key="sdk_...")

# This trace auto-registers the agent in your inventory
with client.trace("customer-support-agent", agent_version="2.0") as trace:
    with trace.span("gpt-4o-chat", type=SpanType.LLM) as span:
        # Your LLM call
        span.set_token_usage(prompt=200, completion=150, model="gpt-4o")

    with trace.span("knowledge-base-search", type=SpanType.TOOL) as span:
        # Your tool call
        span.set_output({"results": [...]})

client.shutdown()

After this runs, your inventory will show:

• Agent: customer-support-agent (LangChain / Python)
• Model dependency: gpt-4o (OpenAI)
• Tool dependency: knowledge-base-search
• Status: MONITORED

Provider Inference

NodeLoom infers the AI provider from the model name automatically:

• gpt-4o, o1, o3 → OpenAI
• claude-3.5-sonnet → Anthropic
• gemini-1.5-pro → Google
• mistral-large → Mistral
• command-r → Cohere
• llama-3 → Meta