Agenten & Multi-Agenten

MLflow

mlflow/mlflow

An open-source AI engineering platform for tracing, evaluating, monitoring, optimizing, governing, and deploying agents, LLM applications, and machine-learning models.

★ 27,3KSterne
⑂ 6,1KForks
2089Offene Issues
PythonSprache
Apache-2.0Lizenz
Q@project.QualityScoreRedaktionelle Bewertung

Übersicht

MLflow combines LLM and agent observability, evaluation, prompt management, prompt optimization, an AI Gateway, experiment tracking, model evaluation, model registry, and deployment workflows. The repository identifies Python as its primary language and states that integrations are available for Python, TypeScript/JavaScript, Java, OpenTelemetry, MCP, numerous agent frameworks, and multiple model providers.

Wichtige Funktionen

  • Tracing and observability for LLM applications and agents, built on OpenTelemetry
  • Systematic evaluation with more than 50 stated built-in metrics and LLM judges, plus custom metrics
  • Prompt versioning, testing, deployment, lineage tracking, and optimization
  • AI Gateway with an OpenAI-compatible interface, routing, rate limits, fallbacks, credential management, guardrails, cost controls, and traffic splitting
  • Experiment tracking for models, parameters, metrics, and evaluation results
  • Model evaluation and collaborative model registry
  • Batch and real-time deployment support for Docker, Kubernetes, Azure ML, AWS SageMaker, and other targets
  • Integrations across Python, TypeScript/JavaScript, Java, OpenTelemetry, and MCP
  • Automatic tracing integrations for more than 60 frameworks, as stated by the README

Voraussetzungen, Installation und Schnellstart

The documented fastest-start command is `uvx mlflow@latest agent setup`. To start the server directly, the README uses `uvx mlflow server`. Alternative package-manager installation procedures and production deployment configuration are not included in the supplied excerpt.

Nutzung

Quick start: run `uvx mlflow server`; in Python, import MLflow, call `mlflow.set_tracking_uri("http://localhost:5000")`, and enable OpenAI logging with `mlflow.openai.autolog()`. Run the instrumented application and inspect traces and metrics at `http://localhost:5000`. The supplied example calls the OpenAI Responses API with model `gpt-5.4-mini`. Additional setup, credentials, and provider configuration are not stated in the repository metadata.

Modellkompatibilität und Anwendungsfälle

The README states support for any LLM provider and lists OpenAI, Anthropic, Databricks, Gemini, Amazon Bedrock, LiteLLM, Mistral, xAI/Grok, Ollama, Groq, DeepSeek, Qwen, Moonshot AI, Cohere, BytePlus, Novita AI, FireworksAI, and Together AI. Listed agent-framework integrations include LangChain, LangGraph, OpenAI Agent, DSPy, PydanticAI, Google ADK, CrewAI, LlamaIndex, AutoGen, Semantic Kernel, Haystack, Vercel AI SDK, Spring AI, and others. Compatibility with every model version or provider feature is not stated.

Lizenz- und Risikohinweise

Repository metadata identifies the license as Apache-2.0. License URL: https://api.github.com/licenses/apache-2.0. Users should review the repository license before redistribution or modification.

Veröffentlichung und Wartung

No release version, changelog, or release notes were supplied. The default branch is `master`; the repository was created on 2018-06-05 and is not archived.

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython

RAGFlow

infiniflow/ragflow

★ 86,7KGo