Агенты и мультиагенты

LangWatch

langwatch/langwatch

An open-core platform for evaluating, testing, tracing, and monitoring LLM applications and AI agents before release and in production.

Открыть официальный репозиторий ↗ Главная страница проекта ↗
★ 3,5KЗвёзды
⑂ 345Форки
840Открытые проблемы
TypeScriptЯзык
Apache-2.0Лицензия
Q98Редакционная оценка

Обзор

LangWatch combines agent simulations, offline evaluations, production observability, datasets, prompt optimization, annotations, and an AI gateway. Its tracing layer is based on OpenTelemetry/OTLP. The repository is primarily TypeScript, while the gateway is distributed as a separate Go binary. Deployment options described by the project include LangWatch Cloud, a local Node.js launcher, Docker Compose, Kubernetes with Helm, and on-premises or hybrid configurations.

Основные функции

  • End-to-end AI agent simulations across tools, state, user simulation, and judging
  • OpenTelemetry/OTLP-native tracing
  • Offline evaluations and datasets
  • Prompt and model optimization workflow
  • Production observability for LLM applications and agents
  • Annotations, review queues, and collaborative failure analysis
  • GitHub integration for prompt management
  • AI gateway with virtual keys, hierarchical budgets, inline guardrails, and provider fallback
  • Cloud, local, Docker Compose, Kubernetes/Helm, on-premises, and hybrid deployment options
  • Integrations with listed frameworks, providers, and automation platforms

Требования, установка и быстрый старт

For the documented local launcher, run `npx @langwatch/server`. It stores services, configuration, generated secrets, and related runtimes under `~/.langwatch/`, starts the services, and opens `http://localhost:5560`. For Docker Compose, clone `https://github.com/langwatch/langwatch.git`, enter the `langwatch` directory, copy `langwatch/.env.example` to `langwatch/.env`, and run `docker compose up -d --wait --build`. The documented development path starts Redis, PostgreSQL, and OpenSearch with `docker compose up redis postgres opensearch`, followed in another terminal by `make install` and `make start`.

Использование

For cloud use, create an account at https://app.langwatch.ai, create a project, and obtain an API key. For local use, open http://localhost:5560 after startup, then create a project and API key. Suggested first steps are running an agent simulation, configuring evaluations, sending traces, or connecting through the MCP integration. Optional local environment flags include LANGWATCH_ENABLE_LANGY, LANGWATCH_ENABLE_PRESIDIO, LANGWATCH_ENABLE_LINGUA, and LANGWATCH_ENABLE_LEGACY_EVALUATORS.

Совместимость моделей и варианты использования

The project describes itself as framework- and LLM-provider-agnostic and supports OpenTelemetry-compatible libraries. Explicitly listed providers are OpenAI, Anthropic, Azure, Google Cloud, AWS, Groq, and Ollama. Listed framework integrations include LangChain, LangGraph, Vercel AI SDK, Mastra, CrewAI, and Google ADK. Listed platforms include LangFlow, Flowise, and n8n. Compatibility with specific model versions is not stated in the repository metadata.

Лицензия и примечания о рисках

The repository metadata identifies Apache-2.0. The README describes an open-core split: most of LangWatch is Apache 2.0, while enterprise modules under `langwatch/ee/`, including SCIM, audit logs, and license or billing management, require a commercial license for production use. The TypeScript SDK, Python SDK, and MCP server are described as MIT licensed. Consult the repository's LICENSE.md, NOTICE, and langwatch/ee/LICENSE.md files for the per-folder terms.

Релиз и сопровождение

Not stated in the repository metadata.

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython

RAGFlow

infiniflow/ragflow

★ 86,7KGo