Opengeni

Opengeni is an Apache-2.0 open-source platform for putting AI agents inside a product without building the underlying infrastructure. Developers create sessions, send input and stream events through one API, using a TypeScript SDK, React components or plain HTTP. It is available as a managed cloud and as a self-hosted deployment.

What It Is

Opengeni is an agent execution layer. Your product supplies the interface, and Opengeni runs sessions, tools and sandboxes underneath. Sessions started by your product also appear in the Opengeni app, where teammates can follow and review them.

What the Platform Handles

The homepage lists the pieces it says you don't have to build:

• Durable sessions that resume after a worker restart or a closed tab.
• Isolated sandboxes for running agent code away from secrets.
• Per-user credential handling, including OAuth tokens and scoped short-lived tokens.
• Tools and MCP, including installing tools from an OpenAPI spec.
• Memory learned from past sessions.
• Multi-tenancy with row-level security.
• Model swapping across OpenAI, Azure OpenAI, OpenRouter or any OpenAI-compatible endpoint.

Embedding in Your Product

@opengeni/sdk/chat provides a server-side chat handler that maps product, tenant, user and conversation. @opengeni/sdk is a TypeScript client for sessions, workspaces, tools, files and event streams. @opengeni/react offers session hooks, a composer and timeline components, which can be restyled through CSS variables for accent, corners, font and light or dark theme. The backend holds the API key and proxies session routes.

Deployment Options

For local development, install Bun and Docker, add a model key to .env and run bun run dev to start the full stack, including the web app. For production, Opengeni provides a Helm chart for any Kubernetes cluster, with Terraform for AWS, Azure and GCP. Compute targets include sandbox execution and Connected Machines, where you enroll a machine you control and run work where your files and tools already live.