Skip to content
Docs menu

Choose how to run Venom

Compare Venom Cloud, your own cloud and your own servers or an air-gapped site, where the data lives in each, what can leave, and where the install guides are.

View .md

Venom is built as software you run, which Venom can also host for you. You pick where it runs, and that decides where your data lives. You can start in one place and move to another: the endpoint is a URL, and sessions are stored on your machine whichever one served them.

The three ways

Venom Cloud Your cloud Your servers or air-gapped
Who runs the control plane Venom You, on your cloud account You, on hardware you own
Where the workspace data lives In Venom Cloud's database In your database On your servers
Where prompts go The target your routing policy picks The same, from your deployment Your own model servers; nothing else if you add no cloud key
What you set up Sign up and create a key Deploy the cloud app and a database Docker Compose, Kubernetes or the air-gapped bundle
Works with the network unplugged No No, unless the model servers are inside your network Yes, with local models

Venom Cloud

The hosted version. You sign up and get one OpenAI-compatible endpoint, the workspace apps and a console. Workspace data (notes, tasks, people and so on) is stored in Venom Cloud's database. Your GPUs stay yours: a daemon in mesh mode dials out to the cloud and keeps its provider keys and models local, and a request either passes through the cloud in transit or is redirected straight to your daemon with a single-use ticket. Start with the Venom Cloud quickstart.

Your cloud

The same control plane and apps, deployed on a cloud account you own, with your own database. Workspace data never reaches Venom's servers, and the deployment keeps running if you stop paying. Self-hosting Venom Cloud lists the variables and make targets, and Deploy covers the orchestrator and the daemons that serve models for it. Install guides for a one-command Linux server and a one-click AWS launch are being written; until then, this is the path.

Your servers or air-gapped

The orchestrator and console on hardware you control, from one machine to a Kubernetes cluster, in front of vLLM, SGLang, llama.cpp or Ollama. With no cloud key set, no request leaves your network. For a site with no network at all, the air-gapped install is built on a connected machine once and carried across. Guides: Docker Compose, Kubernetes, the desktop daemon, air-gapped.

What leaves

Whichever way you run it, two things can leave your infrastructure, and both are off until you turn them on:

  • Model calls to external providers. A request goes to OpenAI, Anthropic or another provider only when your routing policy allows that provider and you have given it a key. A private route sends nothing outside your own daemons.
  • Update checks. Venom ships with no telemetry. The desktop daemon's auto-update is a dashboard switch, off until you flip it, and makes one request for a checksum file per check.

Mesh mode only exists when you set VENOM_CLOUD_URL; without it a daemon dials nothing.

Which one to pick

  • You want to be running in five minutes, and you are fine with workspace data in Venom Cloud: Venom Cloud.
  • Your data may not leave your own account or region, but you have a cloud team: your cloud.
  • You have your own GPUs, a compliance boundary, or no network: your servers or air-gapped.

To use only the coding agent, skip all of this: install the venom CLI and point it at a local model.

See every product on the Products page.