Agentes y multiagente

OpenGeni

cloudgeni-ai/opengeni

An open, self-hostable agentic runtime for organizations that provides durable, replayable agent sessions, human approvals, governed credentials and memory, and flexible compute targets including managed sandboxes and enrolled user hardware.

Abrir repositorio oficial ↗ Página de inicio del proyecto ↗
★ 56Estrellas
⑂ 4Bifurcaciones
53Problemas abiertos
TypeScriptIdioma
Apache-2.0Licencia
Q@project.QualityScorePuntuación editorial

Capturas de pantalla del proyecto

Captura de pantalla de OpenGeni Captura de pantalla de OpenGeni

Resumen

OpenGeni is a platform layer designed to make long-running AI agents safe for real organizational work. Extracted from two years of production experience at CloudGeni, it operates on the premise that safe agent adoption at scale is a platform problem. It provides a session-based API for creating, steering, observing, interrupting, and replaying agent runs. The system is agnostic to what the agent does, acting solely as the runtime. It ensures all control plane data, session states, event histories, and audit trails reside within the user's own deployment. OpenGeni supports two co-equal compute targets: managed sandboxes provisioned by the platform, and Connected Machines—computers enrolled by the user that run sessions directly on local hardware with dial-out-only networking.

Características principales

  • Self-hosted Apache-2.0 runtime with control plane, API, and web app.
  • Durable, replayable sessions backed by a Postgres event log.
  • Human approvals gating tool use and structured agent input requests.
  • Governed credentials brokered per session and reviewed agent memories.
  • Co-equal compute targets: managed sandboxes and Connected Machines.
  • Connected Machines with dial-out-only networking, loud consent enrollment, and one-click revocation.
  • Workspace knowledge layer with document upload, indexing, and hybrid/vector/keyword search via pgvector.
  • GitHub App integration for scoped repository access with independent workspace allowlists.
  • Session goals with success criteria, explicit stopping acts, and budget guards.

Requisitos, instalación y guía rápida

Clone the repository and ensure prerequisites (Bun, Docker, OpenAI/Azure OpenAI credentials) are met. Copy `.env.example` to `.env` and configure the required variables, including `OPENGENI_DATABASE_URL`, `OPENGENI_NATS_URL`, `OPENGENI_TEMPORAL_HOST`, `OPENGENI_OPENAI_PROVIDER`, and model credentials. For local S3-compatible storage, configure the MinIO endpoints. For production, configure native provider object storage variables (e.g., `OPENGENI_OBJECT_STORAGE_BACKEND=aws-s3`).

Uso

To start the full local stack, run `bun run dev`. This command installs dependencies, configures `.env`, starts Docker infrastructure, runs migrations, builds the local sandbox image, and starts the API, workers, and web app. Access the web app at `http://127.0.0.1:3000`, API health at `http://127.0.0.1:8000/healthz`, NATS monitor at `http://127.0.0.1:8222`, and MinIO console at `http://127.0.0.1:9001`. For manual startup, run `bun install`, bring up Docker services, run `bun run db:migrate`, build the sandbox image, and start the API, control worker, turn worker, and web app in separate terminals. To enable Connected Machines, set `OPENGENI_SANDBOX_SELFHOSTED_ENABLED=true` and configure the relay, control-plane endpoints, and signing secrets.

Compatibilidad de modelos y casos de uso

OpenAI Agents SDK, OpenAI, Azure OpenAI.

Notas sobre la licencia y los riesgos

Apache-2.0

Editorial verification 2026-08-02: repository URL, owner, description, license and repository statistics were reviewed. License metadata: Apache-2.0. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

Lanzamiento y mantenimiento

Not stated in the repository metadata

Firecrawl

firecrawl/firecrawl

★ 161,1KTypeScript

LangChain

langchain-ai/langchain

★ 143,6KPython

RAGFlow

infiniflow/ragflow

★ 86,7KGo