Resumen
Future AGI combines OpenTelemetry-based observability, evaluation metrics, agent simulations, datasets, guardrails, prompt optimization, and an OpenAI-compatible gateway. It can run as a managed cloud service or as a self-hosted stack. The repository metadata lists Python as the primary language, while the documented runtime also includes a Go gateway, a React/Vite frontend, and supporting data services.
Características principales
- OpenTelemetry-native tracing with instrumentation for more than 50 AI frameworks
- More than 50 evaluation metrics using LLM-as-judge, heuristic, and machine-learning approaches
- Persona-driven, adversarial, edge-case, text, and voice simulations
- 18 built-in guardrail scanners and 15 vendor adapters
- OpenAI-compatible gateway supporting more than 100 providers and 15 routing strategies
- Prompt optimization with six documented algorithms
- Datasets, experiments, dashboards, annotation queues, and prompt management
- Managed cloud and Docker-based self-hosting options
- Interfaces based on OpenTelemetry OTLP, OpenAI-compatible HTTP, and SQL storage
Requisitos, instalación y guía rápida
Uso
Compatibilidad de modelos y casos de uso
The repository describes an OpenAI-compatible gateway with support for more than 100 providers. Named integrations include OpenAI, Anthropic, Google Gemini, Vertex AI, AWS Bedrock, Azure OpenAI, Mistral, Groq, Cohere, Together, Perplexity, OpenRouter, Fireworks, xAI, Replicate, HuggingFace, Ollama, vLLM, LM Studio, TGI, and Llamafile. Framework integrations include LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, Phidata, PydanticAI, Claude SDK, LiteLLM, Haystack, DSPy, Instructor, and Smol-agents. Exact compatibility by model version is not stated in the repository metadata.
Notas sobre la licencia y los riesgos
The repository metadata identifies the project license as Apache-2.0, and the README points to LICENSE and NOTICE. The README separately says ecosystem SDKs may be Apache- or MIT-licensed, so each independently packaged SDK should be checked before use.
Lanzamiento y mantenimiento
The supplied README describes a nightly release for early testing and says a stable version is coming soon. Recently shipped items include a prompt optimization engine, feed clustering, dataset experiment agent runs, simulations from production calls, an agent playground, dashboards, MCP access, annotation queues, and the Agent Command Center. In-progress or planned work includes fuller agent tracing, multimodal support, agent changelog views, queue assignment, computer-use and coding-agent simulations, and scheduled simulations. Specific release version and release date are not stated in the repository metadata.