Дообучение, обучение и данные

Langfuse

langfuse/langfuse

An open-source LLM engineering platform for tracing, evaluating, debugging, and improving AI applications, with prompt management, datasets, metrics, a playground, APIs, and managed or self-hosted deployment options.

★ 32,4KЗвёзды
⑂ 3,5KФорки
716Открытые проблемы
TypeScriptЯзык
NOASSERTIONЛицензия
Q98Редакционная оценка

Обзор

Langfuse helps teams develop and operate LLM applications collaboratively. It records traces for model calls, retrieval, embeddings, agent actions, and other application logic; supports prompt versioning and experimentation; and provides multiple evaluation workflows. The repository is primarily TypeScript and is not archived.

Основные функции

  • LLM application tracing and observability for model calls, retrieval, embeddings, agent actions, logs, and sessions
  • Central prompt management, version control, collaboration, and caching
  • LLM-as-a-judge, code evaluators, user feedback, manual labeling, and custom evaluation pipelines
  • Datasets for test sets, benchmarks, experiments, and pre-deployment testing
  • Playground for iterating on prompts and model configurations
  • Public API with an OpenAPI specification, Postman collection, and Python and JavaScript/TypeScript SDKs
  • Managed Langfuse Cloud and self-hosted deployment options
  • Integrations with OpenTelemetry, OpenAI SDK, LangChain, LlamaIndex, Haystack, LiteLLM, Vercel AI SDK, Mastra, and other AI tooling

Требования, установка и быстрый старт

For local self-hosting, clone https://github.com/langfuse/langfuse, enter the repository directory, and run Docker Compose as documented at https://langfuse.com/self-hosting/local. The quick-start Python client uses the `langfuse` and `openai` packages. Kubernetes, VM, and cloud-specific self-hosting guidance is available through https://langfuse.com/self-hosting. Exact version constraints are not stated in the repository metadata.

Использование

Create a Langfuse Cloud account at https://cloud.langfuse.com/auth/sign-up or deploy Langfuse yourself. Create a project and API credentials, then configure `LANGFUSE_SECRET_KEY`, `LANGFUSE_PUBLIC_KEY`, and `LANGFUSE_BASE_URL`. Instrument application functions with the Python `@observe()` decorator or use a supported automatic integration, execute an LLM call, and inspect the resulting trace in Langfuse. General documentation is available at https://langfuse.com/docs.

Совместимость моделей и варианты использования

The quick start demonstrates OpenAI with `gpt-4o`. The repository also states integration support for LiteLLM, including Azure, OpenAI, Cohere, Anthropic, Ollama, vLLM, SageMaker, Hugging Face, and Replicate, as well as Amazon Bedrock. Langfuse is model-observability infrastructure rather than a model itself. An exhaustive validated model list is not stated in the repository metadata.

Лицензия и примечания о рисках

Repository metadata reports the SPDX value as NOASSERTION and does not provide a license URL. The supplied README says the repository is MIT licensed except for `ee` folders and directs readers to https://github.com/langfuse/langfuse/blob/main/LICENSE and https://langfuse.com/docs/open-source. Review those terms, especially the exceptions, before deployment or redistribution.

Релиз и сопровождение

No specific release version or release notes are included in the supplied record. The repository default branch is `main`, and the metadata reports the last push at 2026-08-03T09:35:12Z. The project changelog is available at https://langfuse.com/changelog.

DeepSpeed

deepspeedai/deepspeed

★ 43KPython

aikit

kaito-project/aikit

★ 535Go

keras

keras-team/keras

★ 64,3KPython