Release Notes
Infrastructure and Product Changes
A concise record of the infrastructure, routing, auth, billing, portal, and operational work already shipped in this repository across the recent main development cycle.
What shipped recently
- •Provider-controller orchestration, warm-pool provisioning, and better provider health / candidate resolution across the compute layer.
- •Broader auth and portal coverage including device login, Google OAuth, password reset, notifications, credentials, and shared loading states.
- •Managed-runtime diagnostics, artifact downloads, billing/admin surfaces, and API key management for operators and developers.
01
Documentation reconciliation for API, MCP, CLI, and integrations
The public docs were reconciled against the current orchestrator routes, CLI command definitions, standalone MCP server, portal routes, managed templates, billing surfaces, and integration examples. The update keeps existing docs routes in place while making the execution lifecycle contract explicit for external builders.
- •Expanded REST API docs for estimate, submit, list, status, runtime, logs, live logs, cancellation, artifacts, artifact download URLs, callbacks, billing, and credential references.
- •Updated MCP docs for hosted Streamable HTTP with OAuth, local stdio API-key compatibility, current hosted tool names, and cost/destructive action warnings.
- •Corrected CLI docs around submit-time estimation, logs, artifacts, provider node commands, and the absence of standalone estimate/cancel commands.
- •Updated portal and integration snippets to use current package names and shipped routes instead of stale examples.
02
Provider controller, warm pools, and routing health
The compute plane now has deeper provider orchestration rather than simple direct dispatch. Recent backend work added a provider-controller service, warm-pool provisioning paths, richer provider health tracking, and better routing visibility across candidate discovery and selection.
- •Implemented provider-controller service flow and integrated warm-pool provisioning.
- •Added provider health and discovery snapshot methods for better routing state inspection.
- •Improved candidate resolution logging and guards against empty candidate sets.
- •Refined routing logic to avoid noisy WARN paths during provisioner-managed execution.
- •Improved provider fallback handling and API error classification around provider unavailability.
03
RunPod, Vast.ai, and managed provider integration work
The provider adapters themselves became more production-aware. Recent work tightened managed fleet interfaces, improved RunPod and Vast.ai behavior, and aligned search, health, and startup behavior with the control plane’s routing expectations.
- •Refactored the managed fleet provider interface and implemented stronger RunPod and Vast.ai provider support.
- •Changed Vast.ai offer search usage and related registry / alias handling.
- •Moved RunPod health checks off account-balance-style checks and onto explicit health-check behavior.
- •Improved startup probes, warm-pool logging, and traceability for managed launches.
04
Managed runtime diagnostics and launch failure visibility
The platform now does a better job explaining why a managed run failed instead of simply showing a stuck or opaque launch. Runtime callbacks, probe failures, and provisioning edge cases received targeted diagnostics and test coverage.
- •Added startup probe failure diagnostics and improved related runtime logging.
- •Implemented managed runtime callback diagnostics and enhanced job-log handling around runtime events.
- •Fixed startup probe failure reason handling and strengthened related assertions.
- •Adjusted scheduling logic to prevent premature requeueing before managed provisioning metadata is ready.
05
Managed template launches
Managed app templates now have a clearer browser workflow from launch to dashboard readiness. Users can launch curated workspaces, track startup progress, reopen session history, and see actionable failures when provider startup does not complete.
- •Added the /portal/templates session browser for active and past managed app launches.
- •Added per-session progress pages with lifecycle activity, linked job state, service readiness, runtime, and cost.
- •Improved dashboard proxy behavior for app assets and app-local /api and /views requests.
- •Improved managed capacity and startup failure visibility, including Docker Hub pull-rate-limit failures.
06
Auth flows, portal surfaces, and operational UX
A large share of recent product work happened in the user and operator surfaces around auth, credentials, logs, notifications, and shared portal polish. This brought more of the control plane into the UI instead of leaving it implicit.
- •Added device login page and QR-code authentication flow.
- •Added Google OAuth, password reset, redirect-path handling, and stronger auth page UX.
- •Introduced shared execution loading states across auth, admin, and portal route handshakes.
- •Added notifications, credentials management, artifact download surfaces, and improved logs integration in the portal.
- •Added onboarding tour / intro guidance for portal navigation and functionality.
07
Billing, API keys, developer controls, and operator tooling
Operator and developer workflows gained more real control surfaces. Billing pages, account tooling, API keys, and control-plane UI components were expanded so more day-to-day operations are visible in the product itself.
- •Added developer controls and account-management APIs.
- •Added API key management for listing, creating, and rotating keys, then removed earlier mock key plumbing.
- •Enhanced billing pages with stronger activity metrics and clearer USD formatting.
- •Added connection-status badges and control-plane components for clearer operator feedback.
- •Improved jobs, dashboard, and settings/admin surfaces with better empty states, cleaner messages, and fewer mock fallbacks.