Laminar is an open-source all-in-one platform for engineering best-in-class LLM products.

  • Open-source - Laminar is fully open-source, free to use and easy to self-host. Check our Github repo to learn more. We also provide a managed version, for high-volume production use cases.

  • All-in-one - Laminar provides a full suite of tools for LLM engineering, from observability, evaluations and datasets to prompt chain management, data labeling, and more.

  • Engineering - building software with LLMs is essentially a form of test-driven development, Laminar facilitates this process by seamlessly integrating observability, evaluations, and data labeling.

​
Tracing

  • Laminar tracing is framework-agnostic and based on OpenTelemetry.
  • Just add 2 lines of code and you will be tracing entire execution of your LLM app. Laminar auto-instruments popular LLM SDKs and frameworks, such as OpenAI, Anthropic, Langchain, and more.
  • Traces are sent via gRPC and processed by the Rust 🦀 backend, delivering extremely low-latency and high-throughput.
  • Tracing of images is supported out of the box. Audio is coming soon.

Get started with Tracing.

​
Evaluations

​
Offline evaluations

Evaluations are unit tests for your prompts. Without them, any iteration attempt is blind. Laminar gives you powerful tools to build and run evaluations to facilitate the iteration process. Run them from the code, terminal, or as a part of your CI/CD pipeline.

Learn more about offline evaluations.

​
Online evaluations

You can create and configure an LLM-as-a-judge evaluator or a Python scripts evaluators in Laminar UI. Then, Laminar hosts this evaluation logic and runs it with the selection of incoming traces. The results are stored as labels, and you can use them for further analysis, filtering, and improvement.

Learn more about online evaluations.

​
Data labeling and analytics

Laminar provides a UI for labeling and annotating the traces of your LLM applications. Multiple team members can label the same spans to measure human agreement. In addition, our online evaluators can label spans too, so you can measure human alignment of your LLM judge.

You can organize the labeled data into datasets and use them to update your prompts or fine-tune your models.

Learn more about Data labeling and Datasets.

​
Prompt chain management

You can build and host chains of prompts and LLMs and then call them as if each chain was a single function. It is especially useful when you want to experiment with techniques, such as Mixture of Agents and self-reflecting agents, without or before having to manage prompts and model configs in your code.

Learn more about Pipeline builder for prompt chains.

Installation