Changelog

​

Reboot one VM

• POST /v1/orchestrator/nodes/{node_id}/virtual-machines/{vm_id}/reboot restarts one VM and keeps its disk. reboot restarts the operating system and resets the VM if it does not come back; reset power-cycles it at once.
• In Nodes, a VM’s actions menu has Reboot VM and Reset VM, and selecting several VMs offers Reboot and Reset for all of them. Both keep each VM’s disk, and the VM list shows Rebooting until it is back.
• With several VMs in a reservation, the stop button reads Stop all VMs.

​

Reimage a VM from another disk

• POST /v1/orchestrator/nodes/{node_id}/virtual-machines/{vm_id}/reimage rebuilds one VM from a disk you choose without stopping the reservation. It erases the VM’s disk and keeps the VM’s id, hostname, GPUs, and size.
• The new disk boots with the reservation’s SSH keys. If a reimage fails, the reservation stays active and you can reimage again.

​

Pay as you go buys one VM

• Pay as you go buys one VM. Choose 1 GPU, 2 GPUs, 4 GPUs, or 8 GPUs beside the GPU type. A size with no VM available is disabled.

​

Spot VMs come in 1, 2, 4, or 8 GPUs

• Pay as you go lets you choose 1, 2, 4, or 8 GPUs per VM. A smaller VM gets a matching share of its node’s CPUs, memory, and disk.
• Some capacity is sold only as 8-GPU VMs, so smaller sizes can have fewer VMs available.
• ornn spot bid --gpus and the MCP ornn_spot_bid tool accept 1, 2, 4, or 8 (CLI 1.0.49).

​

Ornn reboots a VM whose guest agent stops

• Ornn manages VMs through qemu-guest-agent. It now checks the agent every 5 minutes; if the agent does not answer two checks in a row, Ornn reboots the VM to restore it (a normal reboot, then a hard reset if needed). A reboot stops anything running on the VM, so keep qemu-guest-agent running.

​

Disks replace VM images

• VMs now boot disks. Images in the sidebar is now Disks.
• Create disk picks a base image plus optional Python, packages, CUDA, NVIDIA driver, and DCGM, offering only versions that base image’s Ubuntu release can install. NVIDIA driver choices are 580 (proprietary), 580 (open), and 595 (open). The first VM that launches the disk builds it once; later launches boot the built disk with that software already installed.
• Upload disk adds your own qcow2 from a file or a public URL.
• A created disk builds on its first VM launch, then shows Building, Ready, or Failed. A failed build shows its reason; create the disk again to retry.
• Spot VMs start from your organization’s newest disk, or the newest Ornn disk.
• The CLI and MCP launch requests use disk_id (CLI 1.0.45).

​

VMs can join Kubernetes and Slurm clusters

• Running VMs can join a Kubernetes or Slurm cluster as workers, alone or alongside nodes. Pick them under Virtual machines when you create a cluster or use Add nodes, or pass --vms to ornn launch (CLI 1.0.44).
• While a VM is in a cluster, its SSH keys are set aside and open sessions end. Its disk and data stay, and its keys come back when it leaves.
• Ornn no longer keeps its own login inside your VMs. It reaches a VM through the QEMU guest agent, so custom disks must include qemu-guest-agent.

​

Balanced sidebar spacing

​

Prices are what you pay

• Spot, the Market, and Relist show every price with Ornn’s fees already included. An ask is what you pay to take it; a bid is what a seller receives for filling it. The purchase panel shows one total, and a spot bid shows the price per GPU-hour it pays.
• A limit bid’s price is the most you pay per GPU-hour.
• Relist’s Take the current bid first shows that bid as what you receive, and your asking price is what you receive for each GPU-hour sold.
• Reservations in Nodes show the price you pay, or, on capacity you relisted, the price you receive.
• ornn fees show and the ornn_fees_show MCP tool are removed.

​

Buy relisted capacity at its asking price

• Market purchases replace offers on relisted capacity. Each relist has an asking price. Drag across the time you want and buy it at that price; there is no seller acceptance. One drag can cross reserved time and several relists of the same nodes, which now share one color-coded bar. The purchase panel lists each stretch with its price, the time it skips, and the Ornn trading fee you pay on top.
• Market purchases are paid up front: Ornn credits first, then your saved card, or by wire at $10,000 or more. With no card saved and too few credits, Buy takes you to Billing first. Spot bids now do the same.
• Relist sets an asking price instead of a spot price. A relist can no longer overlap another live relist of the same nodes. Sellers are credited for each sold window when it ends.
• The CLI adds ornn schedules buy and renames the relist price flag to --ask-price-per-gpu-hour-micro-usd; MCP adds ornn_schedules_purchase. Reserve keeps its offers.

​

Reserve, Market, and Spot

• Reserve lists reserved capacity for approved organizations. Post an offer for any window of a listing; the seller accepts. Listings show their down payment, scaled to your window.
• Market lists capacity other customers relisted. Every signed-in organization can drag a window on a listing’s bar and offer its own price, with no down payment.
• Spot rents 1 to 8 GPUs on one node by the hour from a live order book, prepaid from Ornn credits and then your saved card.

​

VM isolation and security updates

​

SSH Keys in the sidebar

​

Storage uploads

​

Image profiles

​

Horizontal Pod Autoscaler

​

Customer workflow paths

​

CLI 0.2.19 — safer terminal workflows

• Destructive cluster, network, and storage commands now confirm before they run. On a terminal they prompt; with --json or in CI they require --yes.
• ornn listings list --operator is removed — Ornn does not expose supplier identity to customers. Filter with --facility.
• Topic help, API errors, wait progress, paginated invoices, TTY tables, and opt-out color make interactive and scripted output easier to trust.
• Guided creation, exact machine hostnames, and the 48-hour capacity forecast shorten common reservation-to-access workflows.

​

Cluster Observability

​

MCP customer paths

​

Interactive SSH only

​

CLI 0.2.2 — observability latest and tail

• ornn telemetry latest / ornn logs latest read a one-shot window or log page. ornn telemetry tail / ornn logs tail follow the same live Observability subscribe feed as the console graphs. MCP exposes telemetry_latest, telemetry_tail, logs_latest, and logs_tail.
• Interactive SSH is ornn ssh <node-id> (add --identity-file for your PEM). There is no host-snapshot or nodes console verb.

​

CLI 0.2.0 and MCP name lockstep

• ornn update installs the latest @ornncompute/cli from npm when a newer release is published.
• Prefer ornn slurm and ornn kubernetes to launch or tear down those controllers. Launch is blocked only when the reservation’s nodes already belong to the other controller.
• MCP session tools are now identity_whoami and identity_status (replacing ornn_whoami and ornn_status). Reconnect the agent if your client cached the old names.

​

ACH (US bank account) checkout is live

• Card settles in seconds — your bid or reservation moves forward immediately.
• ACH typically settles in 1–3 US business days. Your invoice sits in a processing state until the transfer clears; you don’t need to pay again.

​

Shorter Quick Connect SSH usernames

ssh <name>@<ip>

​

Docs visual overhaul

• Banner artwork on every card, plus a lifecycle illustration on the introduction.
• New diagrams: bid lifecycle (since retired) and access modes.
• Refreshed theme and content across all guides.

​

Docs refresh

• Repositioned Ornn as a GPU omnicloud across the docs.
• Documented the full bid lifecycle: Pending deposit → Active → Accepted → Reservation created, plus Rejected, Withdrawn, and Expired.
• Clarified that checkout / down-payment happens before a bid is reviewable.
• Updated Account to reflect four tabs: Profile, Billing, Settings, and SSH keys.
• Added an Access guide with SSH config, port forwarding, data persistence, and troubleshooting.
• Added Manage SSH keys and Resale market guides.
• Added the Ornn Compute CLI reference.