graphsignal 0.19.6

Graphsignal SDK for Python

Graphsignal: Inference Observability

Graphsignal is an inference observability platform that helps developers accelerate and troubleshoot AI systems. It provides essential visibility across the inference stack, including:

• Continuous, high-resolution profiling timelines exposing operation durations and resource utilization across inference workloads.
• LLM generation tracing with per-step timing, token throughput, and latency breakdowns for major inference frameworks.
• System-level metrics for inference engines and hardware (CPU, GPU, accelerators).
• Error monitoring for device-level failures, runtime exceptions, and inference errors.
• AI debugging to explain performance data and errors, identify bottlenecks, and recommend optimizations across the inference stack.

Learn more at graphsignal.com.

Install

Install the Graphsignal library.

pip install -U graphsignal

GPU profiling (Linux): For CUPTI-based GPU profiling, install the extra matching your CUDA version: pip install graphsignal[cu12] (CUDA 12.x) or pip install graphsignal[cu13] (CUDA 13.x).

Configure

Configure the Graphsignal SDK by specifying your API key directly or via the GRAPHSIGNAL_API_KEY environment variable.

import graphsignal

graphsignal.configure(api_key='my-api-key')
# or pass the API key via the GRAPHSIGNAL_API_KEY environment variable

See configure() API docs for all configuration parameters.

To get an API key, sign up for a free account at graphsignal.com. The key can then be found in your account's Settings / API Keys page.

Alternatively, you can use the Graphsignal runner when running your application. Pass the API key via the GRAPHSIGNAL_API_KEY environment variable.

graphsignal-run <my-app>

Integrate

Graphsignal works automatically and also provides an API for manual tracing and profiling. It captures performance profiles, traces, metrics, and errors. All insights are available for analysis at app.graphsignal.com.

Refer to the guides below for detailed information on:

• Manual Tracing
• Manual Profiling
• Using Tags

See integration documentation for libraries and inference engines:

• PyTorch
• vLLM
• SGLang

See the API reference for complete documentation.

Analyze

Log in to Graphsignal to monitor and analyze your application.

Debug with AI

Install the Graphsignal skill to let your AI coding agent (Cursor, Claude Code, Codex, or Gemini) fetch and analyze debug context directly from your IDE. See AI Debugging for setup instructions.

Overhead

The SDK has no impact on production performance that needs to be considered. Profiling uses low-overhead profiling APIs, so it runs with minimal impact on your code. Tracing is lightweight—overhead per trace is typically under 100 microseconds.

Security and Privacy

The Graphsignal SDK only establishes outbound connections to api.graphsignal.com to send data; inbound connections or commands are not possible.

Content and sensitive information, such as prompts and completions, are not recorded.

Troubleshooting

To enable debug logging, add debug_mode=True to configure(). If the debug log doesn’t provide hints for resolving the issue, report it to our support team via your account.

In case of connection issues, please make sure outgoing connections to https://api.graphsignal.com are allowed.