Skip to main content

2026 Releases

Releases marked with "Please update" include important client-side changes — update by running the Windows installer or the Linux/Mac installation scripts.

v0.58.0 (2026-05-15)​

Please update clients.

Rift Server​

  • Smarter NUMA-aware GPU placement for Pinned VMs, with more reliable multi-GPU launches on multi-NUMA hosts.
  • Better vCPU and memory placement across GPU-local NUMA nodes.
  • Prevent active VMs from causing overcommit in hugepage placement checks.
  • Clearer errors when hugepage or NUMA capacity is insufficient.

Rift Desktop​

  • Pinned VMs now expose real per-GPU NUMA affinity inside the guest. Tools like nvidia-smi topo -m, NCCL, CUDA placement helpers, and /sys/bus/pci/devices/*/numa_node can now report correct topology.
  • Keep the legacy flat PCIe layout where required, such as consumer GPUs or hosts with incomplete NUMA metadata.

v0.57.3 (2026-04-30)​

Minor release. We recommend updating the CLI as it had a 2FA authentication bug.

Rift Server​

  • Add POST /api/v1/instances/pause and POST /api/v1/instances/resume endpoints. Pausing freezes vCPU execution; resources stay allocated and billing continues.
  • Add POST /api/v1/nodes/set_marketplace_listing to hide/show a node on the marketplace. Rejects draining nodes with 400.

Rift Desktop​

  • Fix sriov-manage and other PCI driver tools failing with "No such device" on GPUs that were previously passed through to a VM. Unbinding a device now clears its driver_override, so the kernel no longer pins the device to vfio-pci after the VM is torn down.
  • Fix VM creation failing with "Unable to find any usable hugetlbfs mount for 1048576 KiB" when the host has multiple hugetlbfs mounts. The provider now selects the mount whose pagesize matches 1G instead of returning the first hugetlbfs entry in /proc/mounts.

Rift CLI​

  • Add rift virt pause <id_or_name> and rift virt resume <id_or_name> subcommands.
  • Fix error decoding response body when signing in as a user with TOTP 2FA enabled. The CLI now detects the TOTP challenge response from /auth/login, prompts for the 6-digit code, and exchanges the partial token for a full JWT via /auth/totp/complete.
  • Cache the JWT after a successful login so subsequent commands don't re-prompt for a 2FA code on every invocation. The cached token is reused until it expires (server-issued lifetime: 30 days), at which point the CLI silently falls back to a fresh email+password (+TOTP) login. Env-var auth (CLOUDRIFT_USER/CLOUDRIFT_PASSWORD) keeps current behavior and is unaffected.

v0.57.2 (2026-04-24)​

Minor release, no update necessary.

Rift Server​

  • Add admin-only POST /api/v1/users/delete for user erasure. Pre-flight refuses with 409 on provider roots, node owners, or users with active executors. Supports dry_run=true. Hard-deletes credentials and PII-bearing records; anonymizes the user link on retained fraud, payout transaction, and team records so the financial/audit trail survives.
  • Add POST /api/v1/nodes/delete endpoint for admins to permanently delete nodes.
  • Track usage time on billing-exempt rentals so the UI can display runtime for exempt instances and volumes. No transactions are created and balances are still untouched — only the tracker's usage field is advanced by the periodic billing job.
  • Add time-range filtering (from/to) to POST /api/v2/account/transactions/list endpoint. Defaults to last 30 days.

v0.57.1 (2026-04-10)​

Minor release, no update necessary.

Rift Server​

  • Add human-readable instance names (e.g. abaft-yak-4159) to instances. Names are auto-generated on rent if not provided via the name field in the rent request. Exposed in instance list responses as instance_name and in usage transactions as resource_name.
  • Add resource_type (Compute or Volume) and resource_name fields to usage transactions in POST /api/v1/account/transactions/list and POST /api/v2/account/transactions/list.
  • Make gateway optional in POST /api/v1/network/add request. Networks can now be created without specifying a gateway IP.
  • Add billing_exempt flag for admin users to rent instances and create volumes without charges. Intended for internal testing of rentals. When enabled, no use transactions are created for the renter and no payouts are generated for the provider.

Rift CLI​

  • Add --billing-exempt flag to rift instances rent command for admin users.

v0.57.0 (2026-04-01)​

Please update clients.

Rift Server​

  • Extend POST /api/v1/auth/me with id, provider, totp_enabled fields.
  • Add created_at, ssh_key_auth, and user_id fields to instance list response.
  • Team search with empty query returns most recently created teams.
  • [BREAKING] Remove the pat field from /api/v1/auth/login and /api/v1/auth/totp/complete responses. Clients now receive only the JWT token and should not expect a PAT after login.
  • [BREAKING] New users are no longer created with a personal access token automatically, so password login no longer provisions a reusable PAT for fresh accounts.
  • Add team_name field to POST /api/v1/network/list response.
  • Add mig_gpu_mask field to POST /api/v1/nodes/list response.

v0.56.1 (2026-03-25)​

Minor release, no update necessary.

Rift Server​

  • Extend node list responses with proxy connection details (proxy_user, proxy_port) alongside existing SSH connection info.

Rift CLI​

  • Utilize new proxy connection details in rift node ssh to construct SSH connection strings.

v0.56.0 (2026-03-23)​

Please update clients.

Rift Server​

  • Remove node_mode field from reservation list response. Instance mode is only relevant for active executors, not reservations.
  • Add team invite by email: when adding a member by email who doesn't have an account, an invite is created and an email notification is sent. The user is automatically added to the team upon registration.
  • [BREAKING] PUT /api/v1/teams/{team_id}/members/add request version bumped to v056; response is now wrapped in {"Member": ...} (existing user added) or {"Invite": ...} (invite created for non-existent user). Old clients will receive an error indicating the new response format.
  • [BREAKING] POST /api/v1/teams/{team_id}/members/remove request changed to use selector: {"selector": {"ById": "uuid"}} (remove member) or {"selector": {"ByEmail": "email"}} (cancel invite). Returns 204 No Content.
  • [BREAKING] PUT /api/v1/teams/{team_id}/members/update request changed to use selector: {"selector": {"ById": "uuid"}, "role": "..."} (update member role) or {"selector": {"ByEmail": "email"}, "role": "..."} (update invite role).
  • GET /api/v1/teams/{team_id} response now includes invites array with pending team invites.
  • Fix signup bonus and team invite processing not running when RIFT_SERVER_REGISTRATION_DEFAULT_ACTIVE=true.
  • Add POST /api/v1/instances/metrics endpoint for fetching GPU metrics (utilization, temperature, VRAM, power) scoped to a specific instance. Filters node-level GPU metrics by the instance's allocated gpu_mask.
  • Add TOTP-based two-factor authentication (Google Authenticator) for CloudRift accounts. See Two-Factor Authentication for details.
    • POST /api/v1/auth/totp/setup — Generate a TOTP secret and provisioning URI for scanning with an authenticator app.
    • POST /api/v1/auth/totp/confirm — Verify the first TOTP code and activate 2FA on the account.
    • POST /api/v1/auth/totp/disable — Verify a TOTP code and disable 2FA on the account.
    • POST /api/v1/auth/totp/complete — Complete login when 2FA is enabled, exchanging a partial token and OTP code for a full session token.
    • POST /api/v1/auth/login now returns a partial_token instead of a session token when 2FA is enabled on the account.

Rift Desktop​

  • Fix VM boot hang on hosts with newer NVIDIA GPUs (e.g. RTX PRO 6000) caused by OVMF firmware failing to initialize GPU Option ROM.

v0.55.4 (2026-03-11)​

Minor release, no update necessary.

Rift Server​

  • Delete volumes when user balance is depleted.
  • Fix manually disabled GPUs being re-enabled when a node re-registers.

Rift Desktop​

  • Fix potential deadlock during PCI device scanning by moving blocking operations off async worker threads and parallelizing per-GPU setpci probing.

v0.55.3 (2026-03-09)​

Minor release, no update necessary.

Rift Server​

  • Remove unnecessary RwLock around the database client, reducing contention during concurrent operations like node registration.
  • Add /api/v1/users/list endpoint for admin/provider user management with ById, ByProvider, ByMinSpend, and ByActiveExecutors selectors and optional with_account_info.
  • Add /api/v1/users/search endpoint for admin/provider to search users by name or email substring.
  • Add /api/v1/users/send_message endpoint for admin/provider to email selected users.
  • Add /api/v1/users/add endpoint for admin/provider to create users with password-reset-based onboarding. Financial settings (credit limit, initial balance, unlimited) are admin-only.
  • Add /api/v1/account/update_credit_limit endpoint with ByUserIds/ByTeamIds target selector (admin-only).
  • Extend /api/v1/account/transactions/create/external with ByUserIds/ByTeamIds target selectors for admin to add or remove funds.
  • Extend /api/v1/teams/list with ById, ByProvider, ByMinSpend, and ByActiveExecutors selectors, optional with_members/with_account_info.
  • Add /api/v1/teams/search endpoint for admin/provider to search teams by name substring.
  • Extend /api/v1/instances/list with ByUserId selector for admin/provider to list a specific user's instances. Exclude inactive instances by default for all selectors except ByStatus and ByID.
  • Add team authorization for admin and provider root users in /api/v1/instances/list. Provider results are scoped to their own nodes.
  • Track last_login_at on user login.

Rift Desktop​

  • Added support for MIG profiles when renting whole GPUs. This eliminates the need for driver switching, making whole-GPU rentals more seamless.

v0.55.2 (2026-03-09)​

Minor release, no update necessary.

Rift Server​

  • Add /api/v1/nodes/metrics/list API for receiving GPU/MIG instance metrics.

Rift Desktop​

  • Launch dcgm-exporter Docker container automatically when GPUs use NVIDIA drivers; stop it before switching to VFIO to prevent device handle conflicts.

v0.55.1 (2026-03-06)​

Minor release, no update necessary.

Rift Server​

  • Fix instance reporting as Container instead of VirtualMachine during provisioning.
  • Add manual GPU enable/disable flag to exclude flaky GPUs from allocation without draining the node.
  • Split erase_grace_period into two settings: erase_grace_period (default 0s, for explicit destroys) and erase_stray_grace_period (default 72h, for stray VM cleanup). Prevents orphan VMs from filling disk during stray sync.
  • Add TC QoS prioritization on macvtap interfaces to prevent bulk VM downloads from starving SSH and ICMP traffic.

Rift Desktop​

  • Add TC QoS prioritization on macvtap interfaces to prevent bulk VM downloads from starving SSH and ICMP traffic.
  • Handle split erase grace period settings (erase_grace_period vs erase_stray_grace_period) when destroying VMs.

v0.55.0 (2026-03-06)​

Please update clients.

Rift Server​

  • Fix ById instance selector returning empty results for instances rented under a team.
  • Add SelectorScope to ByStatus and ByClusterName instance selectors (Personal, Teams, PersonalAndTeams, All). Team API keys auto-resolve to their team when no scope is specified.
  • Remove deprecated ByTeamId from v055 instance selectors. v045 ByTeamId is still accepted for backward compatibility.
  • Add AMD GPU support: PCI detection, Docker runtime integration, and ROCm verification for AMD Instinct GPUs (MI300X/MI350X).
  • Add AMD SR-IOV (GIM) GPU passthrough for VMs with multi-GPU VF assignment support.
  • Detect unresponsive GPUs via PCI config space status register read.
  • Add ROCm VM image build scripts (Ubuntu Noble, ROCm 7.2, HWE kernel).
  • Fix stray VM detection for domains with legacy metadata missing gpu_addresses field.
  • Fix metadata deserialization failure (missing field type) for legacy domains with older disk source format.
  • Add automatic domain cache reconciliation to prevent stale libvirt handle errors.
  • Add raw XML fallback for executor_id extraction when metadata deserialization fails.
  • Add support for NVIDIA AI Cloud token when provisioning a VM.
  • Add POST /mcp endpoint exposing CloudRift REST API as MCP tools for LLM integration (Claude Code, Claude Desktop, etc.). Supported tools: list_instance_types, rent_instance, list_instances, terminate_instances, stop_instances, start_instances, get_account_info.
  • Refresh domain cache from libvirt at the start of server VM sync.

Rift CLI​

  • Add rift instance-type list command to list available instance types with availability info, with optional --service and --datacenter filters.
  • Add rift instance rent command to rent VM or Docker instances via the public API with support for VM (via --image) and Docker (via --docker-image) modes.

v0.54.0 (2026-02-27)​

Please update clients.

Rift Server​

  • Implement automatic periodic synchronization of Virtual Machines to clean up inactive or stray resources.
  • Volumes can now be accessible from multiple datacenters when the underlying storage cluster is shared. The /public/volumes/list, /public/volumes/create, and /public/team-volumes/* endpoints now return a new datacenters field — an array of all datacenter names where the volume is available. The existing datacenter_name field remains unchanged for backward compatibility.
  • Optimize server by caching immutable node hardware configuration.
  • The /api/v1/auth/me endpoint now supports team API key authentication. Previously, this endpoint only worked with user tokens and user API keys.

Rift Desktop​

  • Implement SyncVirtMachines command to synchronize local VMs with the server, ensuring stray VMs are removed and inactive executors are stopped.
  • Create and destroy default compute instances for MIG devices.
  • Added support for MIG devices in the Rift Desktop client.

v0.53.3 (2026-02-19)​

Minor release, no update necessary.

Rift Server​

  • Optimize server by removing the hot path of fetching node data when not necessary.
  • Optimize server by removing the hot path of fetching instance data when not necessary.

v0.53.2 (2026-02-16)​

Minor release, no update necessary.

Rift Server​

  • Allow users and teams to create and manage their own recipes for virtual machines and containers.

Rift Desktop​

  • NVIDIA / VFIO driver switch hardening. Check for active Docker processes that use GPU before unloading driver to prevent soft-lock.

v0.53.1 (2026-02-06)​

Please update clients.

Rift Server​

  • Fixed a race condition with FRP tunneling that led to occasional failure when allocating ports in tunneling scenarios.

Rift Desktop​

  • Restored an enlarged PCI hole 64 setting to fix virtualization for 8×L40S VMs.
  • Added a check that the NVIDIA driver works before starting a container on the selected GPUs. This prevents sporadic container startup failures caused by uninitialized drivers.
  • Fixed driver switch issues that caused occasional problems on nodes configured for dual container/VM passthrough.
  • Manually start/stop the nvidia_persistenced daemon when switching drivers.
  • Fixed a bug that caused not all GPU functions (video/audio) to be switched.

v0.53.0 (2026-01-30)​

Please update clients.

Rift Server​

  • Fixed the logic for retrieving VM configurations to correctly handle optional datacenter IDs. The system now first attempts to find a configuration specific to that datacenter, falling back to a global configuration associated with the compute provider.
  • When synchronizing VM images on nodes, only images relevant for the node's hardware configuration are selected. For example, nodes without a GPU will only receive VM images that do not require a GPU.

Rift Desktop​

  • Switched to EFI BIOS to reduce GPU passthrough issues and improve performance.

v0.52.4 (2026-01-23)​

Minor release, no update necessary.

Rift Server​

  • Added node drain management APIs for provider maintenance:
    • POST /api/v1/nodes/drain — Start draining a node (hides from marketplace, notifies users).
    • POST /api/v1/nodes/cancel_drain — Cancel an active drain.
    • POST /api/v1/nodes/force_drain — Immediately terminate all instances.
    • POST /api/v1/nodes/resume — Bring a drained node back online.
  • Fixed an issue where Docker containers on the same node with the same port requests would conflict. Containers now bind correctly to their allocated IP, allowing multiple containers to expose the same port.

Rift Desktop​

  • Docker containers are now terminated gracefully to avoid putting the GPU into an invalid state.
  • Fixed an issue where Docker instances with volumes attached were failing to initialize.

v0.52.3 (2026-01-16)​

Minor release, no update necessary.

Rift Server​

  • Send low balance alert emails to team owners when their account balance drops below a predefined threshold.
  • No usage is recorded for failed instance creation attempts. When instance creation fails, the system no longer logs usage time or incurs charges.

v0.52.2 (2026-01-08)​

Minor release. We recommend updating the CLI as it had an authentication bug.

Rift Server​

  • Ensured that Swagger and RapiDoc use the API key authentication scheme for instance endpoints. This fixes instance API calls from Swagger/RapiDoc when using an API key.

Rift Desktop​

  • Fixed an issue where virtual machines were missing their IDs after being destroyed or restarted. The system now correctly retrieves and maintains VM IDs.
  • Added support for additional disks in raw format. These disks can be attached to virtual machines for greater storage flexibility.

Rift CLI​

  • Fixed an authentication bug that caused all rift commands to fail when username/password authentication was used.