Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

Zymtrace

zymtrace.com

  • Free plan
  • Linux
  • Web
  • API
  • Self-hosted
7.4 Club score Visit zymtrace.com

What Zymtrace is

Zymtrace is a CPU and GPU profiling platform for finding performance bottlenecks in AI training and inference. Its whole-system CPU profiler supports languages such as Python, Java, Go, Rust, C/C++, Node.js, Ruby, .NET, PHP, Perl, Zig, and Scala. The GPU profiler links CPU and GPU execution and examines CUDA kernels, GPU instructions, and stall reasons. GPU monitoring includes utilization, memory, power, temperature, SM efficiency, SM occupancy, Tensor Core utilization, and data-transfer rates. Listed inference engines include vLLM, Ollama, llama.cpp, and SGLang.

Zymtrace runs as self-hosted software, with the maker stating that data stays inside the customer’s infrastructure. Its air-gapped deployment performs processing and analysis locally. Each instance exposes a REST API, and profiling data can be made available through MCP for AI agents. OpenTelemetry resource attributes are supported, with export in the standard OpenTelemetry format. The product currently runs on Linux because it relies heavily on eBPF; Windows is not supported. The agent targets 1% CPU usage and under 250 MB of RAM. Basic is free forever for CPU profiling on up to five concurrent machines, with GPU profiling available as a trial. Pro and Enterprise add broader deployment, support, and capacity options.

Who it suits

Zymtrace fits engineering and AI teams profiling CPU and GPU workloads on Linux machines. It is also suited to organizations that require self-hosting, local or air-gapped processing, OpenTelemetry export, or interfaces for AI agents.

Good to know, and what to watch

  • Profiles CPU workloads across many programming languages
  • Correlates CPU and GPU execution
  • Monitors detailed GPU utilization and efficiency metrics
  • Supports self-hosted and air-gapped deployments
  • Provides REST API, MCP, and OpenTelemetry export
  • Offers a free Basic plan forever
  • Currently limited to Linux machines
  • Windows is not supported
  • Basic limits CPU profiling to five concurrent machines
  • Continuous alerts are not available

Enterprise

—

Pro

—

Basic

Free

Free plan

Getting started

  1. Open the Zymtrace website
  2. Choose the self-hosted deployment route
  3. Start with the free Basic plan or request Pro or Enterprise pricing
  4. Install Zymtrace on Linux machines
  5. Use the instance REST API or MCP exposure for integrations
  6. Configure OpenTelemetry attributes or export

Limits to check

Basic is free forever but limits CPU profiling to up to five concurrent machines; GPU profiling is available as a trial. Zymtrace currently supports Linux only, and Windows is not supported.

The verdict

Pick Zymtrace when Linux-based teams need detailed CPU and GPU analysis while keeping profiling data in their own infrastructure. Consider another option if Windows support or continuous alerts is required, or if you need more than five concurrent machines on the free plan.

Questions people ask

Is there a free plan?

Yes. Basic costs nothing and is billed Free Forever, with CPU profiling for up to five concurrent machines and community support.

What does Pro include?

Pro includes CPU and GPU profiling, unlimited machines, air-gap installation support, and dedicated managed hosting availability. Pricing is based on machines and GPUs.

What platforms are supported?

Zymtrace is available through API, Linux, self-hosted deployment, and the web. The maker says it is currently limited to Linux machines.

Does customer data leave the infrastructure?

The plans are self-hosted, and the maker says no data leaves the customer’s infrastructure. The air-gapped option performs processing and analysis locally.

Which inference engines are supported?

The GPU profiling page lists vLLM, Ollama, llama.cpp, and SGLang.

What support comes with each plan?

Basic includes community support. Enterprise lists dedicated Slack and Google Meet support plus custom SLAs.

What the maker publishes

CPU profiling
Its whole-system profiler supports languages including Python, Java, Go, Rust, C/C++, Node.js, Ruby, .NET, PHP, Perl, Zig, and Scala. source
Deployment
The pricing page says all plans are self-hosted and that no data leaves the customer’s infrastructure. source
GPU monitoring
GPU monitoring covers metrics such as utilization, memory, power, temperature, SM efficiency, SM occupancy, Tensor Core utilization, and data transfer rates. source
GPU profiling
The GPU profiler correlates CPU and GPU execution and analyzes CUDA kernels, GPU instructions, and stall reasons. source
Headquarters
Wilmington, Delaware, United States source
Inference engines
The GPU profiling page lists vLLM, Ollama, llama.cpp, and SGLang as supported inference engines. source
Integrations
Zymtrace exposes profiling data through MCP for AI agents and provides a REST API in each instance. source
Observability
Zymtrace supports OpenTelemetry resource attributes and can export profiling data in the standard OpenTelemetry format. source
Operating system
The maker says Zymtrace is currently limited to Linux machines because it relies heavily on eBPF. source
Product limit
The maker says Windows is not currently supported because eBPF is not well supported there. source
Purpose
Zymtrace profiles CPU and GPU workloads to identify performance bottlenecks and optimize AI training and inference. source
Resource footprint
The homepage says the agent targets 1% CPU usage and less than 250 MB of RAM. source
Security and data handling
The maker says all processing and analysis are performed locally in its self-hosted, air-gapped deployment option, with no data leaving the customer’s infrastructure. source
Support
Basic includes community support; Enterprise lists dedicated Slack and Google Meet support and custom SLAs. source

Checked against the maker's pages on 5 Oct 2026.