Hi Product Hunt. For the last two years we’ve been running AI agents against production cloud infrastructure at CloudGeni. What we learned the hard way is that the model was rarely the problem. The problems were operational: a run dying when a browser tab closed, nobody knowing what the agent did at 3am, an agent “helpfully” applying a change no human had looked at, credentials scattered across sandboxes. Safe agent adoption at scale turned out to be a platform problem, not an agent problem. So we pulled the platform out and open-sourced it. That’s OpenGeni. What it is: a self-hostable runtime for long-running agents. It is the runtime, not the agent — you bring the model and the tools, and your product talks to a session API to create, steer, observe, interrupt and replay runs. What’s different - 🧾 Durable, replayable sessions — every event lands in Postgres, so any client can reconnect and see the same history. Live updates stream over SSE and backfill from the log. - ✋ Human control built in — approvals for risky tools, structured input mid-run, and agent memories staged for review before they persist. - 🖥️ Your compute, your choice — run a session in a managed sandbox, or enroll your own laptop, build server or GPU box. Connected Machines dial out only; nothing inbound, no platform-minted credentials on your hardware. - 🎯 Goals with an explicit end — success criteria, budgets, and a session that finishes with goal_complete plus evidence, or goal_pause with a reason. - 🔩 Boring, proven infrastructure — Postgres, Temporal and NATS. Helm and Terraform for Azure, AWS and GCP. TypeScript SDK and React components. Who it’s for: teams embedding agents into their own product who need runs that survive, are auditable and can be interrupted; and platform teams who want agents on their infrastructure, under their controls. Pricing: the runtime is Apache-2.0 and free to self-host. OpenGeni Cloud is the hosted version at model cost + 5% — no seats, no platform fee, no minimum. Enterprise deployments: let’s talk. We are in early public preview, and honestly the feedback is the point of today. What would stop you from putting an agent on real work? What’s missing? Which control would you need before letting one touch production? Tell us here — we’ll be in the comments all day. Try it: git clone https://github.com/Cloudgeni-ai/... && cd opengeni && bun run dev brings up the API, workers, Postgres, Temporal, NATS and the web app with one command. Docs at https://docs.opengeni.ai, hosted app at https://app.opengeni.ai.

Forum Threads

p/opengeni

•

5d ago

Opengeni is getting ready for ProductHunt!

Hey Product Hunt! I'm Davlet from the Opengeni team.

For the past two years we've been running AI agents against real production cloud infrastructure at Cloudgeni. The model was never the hard part. Everything around it was: runs that die when a worker restarts, code that needs somewhere safe to execute, OAuth tokens for every user, tools to wire up, a human sign-off before anything risky, and a clear record of what the agent did and what it cost.

We kept rebuilding that layer, so we turned it into Opengeni.

What you get out of the box:

View all

Opengeni

Ship AI agents within minutes. Infrastructure for Agents

Ship AI agents within minutes. Infrastructure for Agents

Forum Threads

Opengeni is getting ready for ProductHunt!

Forum Threads

Opengeni is getting ready for ProductHunt!