furybot

We do not monitor your workloads

Galtops does not monitor what you do with your models, prompts, or workloads on your hardware. Inference stays on your side of the architecture; we are not in the request path for private deployments.

Pulling a license is commercial and operational control only — whether the organisation is paid up and within its allocation. It is not content surveillance, prompt review, or output inspection.

License to operate (pull)

Deployments poll a license server using STUDIO_LICENSE_URL (recommended host: license.galtops.io; license-server.galtops.com is a documented alias). The control plane path GET /edge/license is the same entitlement surface.

An infrastructure provider (DigitalOcean, RunPod, AWS, bare metal, or self) must be configured on the tenant before management automation — budget apply and related alerts — engages. Without a provider, the deploy still polls, but apply_limits stays off.

Default poll interval is five minutes. The agent caches the last successful refresh locally so a brief outage does not invent a suspension.

Unreachable license server

If the deployment cannot reach the license server (network or DNS — not the same as unpaid), it shows a clear warning and may continue operating for up to thirty days from the last successful license refresh.

After thirty days without a successful refresh, inference is suspended: no GPU or model usage, including chat, the OpenAI-compatible /v1 API, and image jobs, until connectivity is restored and the license is valid.

Unpaid or suspended by the server

When the license server reports that the organisation is suspended or otherwise must refuse inference, the deployment suspends inference promptly — immediately by default, or within an optional courtesy window of at most three days. That courtesy is not the thirty-day unreachable grace; unpaid does not wait thirty days.

What disconnect means

Disconnect means no GPU or model usage until the license is restored. It does not mean an instant wipe of your data.

Self-hosted deployments

On your infrastructure, our software stops serving when the license requires it. We cannot wipe your disks, and we do not operate them as a free forever backup service.

Export support from us is finite: thirty days after inference is suspended. After that window we have no obligation to provide endless export assistance or key support.

If your encryption key is held only on your machine, we never could decrypt your disk. Once the licensed software will not run, recovery is your backups. Fair but firm.

Cloud sold through Galtops

When we deploy for you on a chosen cloud (DigitalOcean, RunPod, AWS, or similar), we operate the hosted lifecycle: suspend inference, then a thirty-day export window, then teardown under this policy and your agreement.

Aggressive short teardown schedules (for example lock at fifteen days and delete at thirty) may apply to demo or trial tenants when disclosed; they are not the default for paying B2B without a written agreement saying so.

Child-safety floor

The child-safety floor is never liftable on any build we ship. It is independent of license status, tenant settings, and experimental mode. A license field cannot turn it off — on Furybot, on a customer deploy, or anywhere else.

Questions

Questions about entitlement, export windows, or a deployment under this policy: use the contact page. Child-safety limits that apply on every build are summarised at what we block on the public demo.