Forge focuses primarily on Forge Voice for home repair shops.
Forge Desk is a separate Forge product that remains available for existing customers. New inquiries here route to /contact.html.
Your dev box. In the cloud. In under sixty seconds.
Your engineers are burning hours on slow laptops, IT tickets for Docker permissions, and security reviews for every new contractor. Forge Desk spins up a full Windows or Linux dev environment in under a minute, pre-loaded with your stack, reachable from any browser or thin client, and wrapped in zero-trust access controls. Ship code, not VPN tickets.
Real hardware. Real throughput.
No nested virt tricks, no shared cores, no oversubscribed noisy neighbors. Forge Desk runs on bare-metal hypervisors with dedicated vCPU reservations, NVMe-backed persistent storage, and 10 Gbps uplinks per node. Your laptop fan stays silent because the work is happening somewhere else.
CPU Tiers
- Dev workstation4 vCPU / 16 GB RAM
- Pro workstation8 vCPU / 32 GB RAM
- Power workstation16 vCPU / 64 GB RAM
- Dedicated reservations, no bursting
- AMD EPYC Genoa / Intel Sapphire Rapids
- Custom tiers on request (up to 64 vCPU)
GPU Tiers
- T4 (16 GB)entry CUDA / light inference
- A10 (24 GB)training, stable diffusion, CAD
- L40S (48 GB)serious ML, batch inference
- GPU burst on Pro tier (hourly)
- Dedicated GPU on Power tier
- CUDA 12.4 + cuDNN pre-installed
Storage + Network
- NVMe SSDup to 1 TB per workstation
- 10 Gbps uplink per node
- Persistent $HOME across reboots
- Hourly snapshots, 7-day rolling
- Private subnet peering (AWS / GCP)
- Optional static egress IP
Everything pre-installed. Nothing to babysit.
Pre-loaded dev tools
Node, Python, Go, Rust, Docker, Postgres, VS Code, JetBrains Toolbox, Git, GitHub CLI, AWS CLI, gcloud, kubectl, Terraform. Opinionated defaults, overridable by team templates.
One-click boot
Click a workspace, get a desktop. 45 to 60 seconds from cold start to a logged-in environment with your dotfiles, SSH keys, and active repos cloned. No Windows update waiting. No reboot roulette.
Zero-trust access
Cloudflare-style identity-aware proxy, SAML or OIDC SSO, device posture checks, just-in-time access grants. No VPN. No open RDP. Every session logged and revocable in one click.
Persistent $HOME
Shut down your workstation Friday, power it on Monday, find everything exactly where you left it. Dotfiles, repos, shell history, open editors. Backed by NVMe, snapshotted hourly.
Team templates
Define a Dockerfile or Nix config once, clone it to every engineer. New hire in Monday morning, coding by 10 AM on the same stack as the senior staff. Template changes reach everyone on next boot.
OS choice
Windows 11 Pro for .NET shops and Unity devs. Ubuntu 24.04 LTS for the sensible majority. Fedora Workstation for the people who like pain. All three ship with matching dev toolchains and VS Code pre-configured.
Built for teams that ship from anywhere.
Client isolation, per project
One workstation per client codebase. No cross-contamination, no accidentally committing Client A secrets into Client B repos. Wind down when the engagement ends.
No laptop shipping
Hire a contractor in Poland on Monday, give them a browser URL, they code by Tuesday. When the contract ends, revoke access and the data never left your perimeter.
GPU sandboxes, on demand
Spin up an L40S workstation for a two-week training run, burn it down after. Pay for the GPU hours you used, not the year-round lease on a depreciating GPU rig in the closet.
Dedicated customer envs
One persistent workstation per enterprise customer, for replicating bugs against their environment. Snapshot it, share it with a teammate, restore it to a known-good state in one click.
Priced per seat. Shut it down on vacation.
Monthly, per engineer. Workstations suspend when idle, so you only pay for what you actually use on Dev tier. Annual pricing available.
- 4 vCPU / 16 GB RAM
- 200 GB NVMe persistent $HOME
- Ubuntu 24.04 LTS only
- 8 hours daily active usage
- Community Discord support
- US region
- 8 vCPU / 32 GB RAM
- 500 GB NVMe persistent $HOME
- Windows 11 Pro, Ubuntu, or Fedora
- GPU burst (T4 / A10, hourly)
- 24/7 availability, no usage caps
- Team templates + shared secrets vault
- SSO (SAML / OIDC)
- US / EU regions, priority email support
- 16 vCPU / 64 GB RAM
- 1 TB NVMe persistent $HOME
- Dedicated A10 GPU (24 GB)
- Always-on, no idle suspension
- Dedicated static egress IP
- Private subnet peering (AWS / GCP)
- US / EU / custom regions
- Priority support, 4-hour response SLA
What dev leads ask first.
How is this different from GitHub Codespaces or Coder?
Codespaces is a browser tab running a container. Great for a quick PR review, painful as a daily driver on anything with a heavy native toolchain or a GPU workload. Coder is self-hosted, which means your platform team is now running a DaaS platform instead of shipping product. Forge Desk is fully managed, gives you a real OS (Windows, Ubuntu, Fedora) on real hardware with a persistent $HOME and GPU options, and costs less than the loaded laptop you would otherwise be buying. If you only need a container, use Codespaces. If you need an actual workstation that boots fast and does not fight you, use Desk.
Can I bring my own image or template?
Yes. Pro and Power tiers support custom Dockerfiles, Nix configs, and qcow2 VM images. Upload once, apply to your whole team. We keep a library of starter templates (Rails, Django, Next.js, FastAPI, Unity, PyTorch, Unreal) you can fork and modify.
What about data egress? Am I going to get surprised on a bill?
No. Standard egress (browser sessions, package downloads, git pushes) is included in the seat price with a 500 GB monthly fair-use cap. If you are pulling multi-TB training datasets every day, talk to us about a custom plan or use Power tier with private subnet peering to your own cloud, which is egress-free.
GPUs are always booked everywhere. Can I actually get one?
T4 and A10 capacity is available in US-East and EU-West on demand today. L40S is by reservation (we will confirm capacity before you commit). H100 is waitlist only, honest about it. If you need guaranteed GPU capacity for a training run, we will hold it for you with a two-week advance reservation.
How is the security model structured?
Zero-trust by default. No VPN, no open inbound ports, no shared admin credentials. Every session is authenticated through your IdP (Okta, Entra ID, Google, JumpCloud), authorized by device posture policy, brokered through an identity-aware proxy, and logged to your SIEM. Workstations are isolated per user, encrypted at rest (AES-256), and snapshotted hourly. HIPAA BAA available on Pro and Power.
What happens when my internet goes down?
Your workstation keeps running, your builds keep building, your long-running training jobs keep training. When you reconnect, your session restores exactly where you left it. That is the hidden win: cloud desktops are actually more resilient to flaky home Wi-Fi than your local laptop is.
Stop waiting on laptops. Boot a dev box that never slows down.
Start a free trial, pick an OS, and be writing code in your first workstation in under five minutes. No credit card, no sales call, no seven-page security questionnaire.