BitPulse

Production-grade LLM observability for GraphBit workflows.

• Zero-config automatic tracing for LLM calls and workflows
• Captures prompts, responses, tokens, latency, errors
• Sends trace data to your observability API endpoint

Project Info

• Python: 3.10–3.13
• License: GraphBit Framework License
• PyPI: bitpulse
• PRs: Welcome via GitHub pull requests

Install

pip install bitpulse

Python: 3.10–3.13

What It Does

BitPulse automatically captures detailed trace data from GraphBit LLM clients and workflows and submits it to your observability endpoint for monitoring, analytics, and debugging.

Key Features

• Automatic tracing for LLM calls and workflows
• Rich metadata: prompts, responses, tokens, latency, finish reasons, errors
• Tool-call detection for agent workflows
• Works with OpenAI and GraphBit internals
• Type-safe models (Pydantic) and robust async I/O

Quick Start

import asyncio, os
from graphbit import LlmClient, LlmConfig
from bitpulse import AutoTracer

async def main():
    tracer = await AutoTracer.create()
    cfg = LlmConfig.openai(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o-mini")
    client = tracer.wrap_client(LlmClient(cfg), cfg)
    resp = await client.complete_full_async("What is GraphBit?", max_tokens=100)
    results = await tracer.send()
    print("sent:", results["sent"], "failed:", results["failed"])

asyncio.run(main())

Configuration

Set environment variables to configure external endpoints:

export BITPULSE_TRACING_API_KEY="your-api-key"
export BITPULSE_TRACEABLE_PROJECT="your-project-name"
# optional
export BITPULSE_TRACING_API_URL="https://your-api-endpoint.com/traces"

Links

• GitHub: InfinitiBit/bitpulse
• Issues: issue tracker

License

GraphBit Framework License. See LICENSE.md in the repository.

Trace Data Format Example

Example JSON payload sent to your observability endpoint:

{
    "tracing_api_key": "your-api-key",
    "traceable_project_name": "your-project",
    "run_name": "LlmClient",
    "run_type": "llm",
    "status": "success",
    "input": "Hello, world!",
    "output": "Hi there!",
    "error": null,
    "start_time": "2025-01-01T00:00:00Z",
    "latency": 123.45,
    "tokens": 50,
    "metadata": {
        "model_name": "gpt-4o-mini",
        "provider": "openai",
        "input_tokens": 10,
        "output_tokens": 40,
        "finish_reason": "stop"
    }
}

API Reference

High-level AutoTracer methods:

• AutoTracer.create(): Initialize tracer
• wrap_client(llm_client, llm_config): Trace LLM client calls
• wrap_executor(executor, llm_config): Trace workflow execution
• send(): Convert and submit captured spans to your endpoint
• export(): Export captured traces locally