Self-hosting: the whole product inside your network

• Code and credentials may not leave the company network at all, by policy or by contract.
• An audit requires the whole path to be under your control: who connects, where sessions run, where the records are stored.
• You already run a VPC and an ops team, and do not want an outside dependency on the critical path.

• The entry point and the console run in your data center or VPC. Clients connect to your address, not ours.
• Connecting one of your machines is still one command, and it still needs no public port on that machine.
• Accounts, tiers and the list of connected servers live in your own database.
• Nothing on the path goes through us, which also means nothing on the path depends on us being up.

Two things do. It is better to read them here than to find them in a review:

• Claude itself talks to Anthropic. Sessions run on your own Claude subscription, so that traffic leaves your network directly. It does not pass through us, but it does go out. Keeping it inside is a question for your egress policy.
• Clients and updates come from us. The apps can be mirrored onto an internal share, but the packages are signed by us. That is the other side of shipping binaries rather than source.

Clients are used on the move: a phone on the train, a laptop at home. So the entry point either has to be reachable from outside, or it sits inside the network and clients join the VPN first.

Both work. The second turns “pick it up anywhere” into “VPN first, then pick it up”. That is a product decision rather than a deployment detail, so settle it before the work starts.

• Binaries and deployment docs. RoamAI is not open source.
• One hands-on setup: standing it up, pointing the clients at it, getting the first session running.
• Upgrades are a new binary plus a note on what changed. When to take them is your call.
• Contract, invoice, and payment by company transfer.

Write to support@roamai.cloud, or add us on WeCom. Answer these three questions in the first message and we can reply with a concrete proposal instead of more questions:

1. roughly how many people;
2. which machines the work should run on: servers you already have, or ones we help you set up;
3. where the entry point sits: reachable from outside, or inside the network only.