Opengeni: Ship AI agents within minutes. Infrastructure for Agents
Opengeni is an Apache-2.0 runtime from CloudGeni, a company that spent two years running agents against production cloud infrastructure. The team says the lesson from that work was that safe agent adoption is a platform problem more than an agent problem. The project exposes a session-based API for creating, steering, observing, interrupting and replaying agent runs, whatever the agent does. Each session can run in a managed sandbox (Docker, Modal or a cloud provider) or on a "Connected Machine" such as a laptop or GPU box. The bundled React app is only one client, and products can call the API directly. Recent repository changes reposition the project toward embedding agents inside other products. That means per-user credentials, memory, multi-tenant workspaces, MCP tools and restylable React components. New docs compare it with LangGraph, CrewAI, Mastra, the Vercel AI SDK, the OpenAI Agents SDK, Letta and Inngest AgentKit. For founders, the main question is build versus adopt. Durable sessions, approval gates and audit trails are costly to build and rarely differentiate a product. A permissive license and a self-host option reduce lock-in. The project is young, though, and several forks and lookalike pages already exist, so check provenance before depending on it.