What's new at Domino

Domino Jobs, including Ray and Spark, now run on Arm compute such as AWS Graviton, Azure Cobalt, Google Axion, and NVIDIA Grace-based GPU systems, on remote data planes as well. Admins add arm64 node pools and Hardware Tiers next to the existing x86 ones, and one Domino Environment serves both.

Why it matters:

• Run Spark and Ray on Graviton, Cobalt, or Axion for better cloud price-performance, or on Grace-based systems when the work needs GPUs.
• Move a Job between Arm and x86 in the same project by picking a different Hardware Tier.
• Review Arm runs with the same logs, results, and artifacts you get from any x86 Job.

Domino Skills now ship as one plugin for Claude Code, Codex, Gemini CLI, Antigravity, and Cursor, with two MCP servers: one starts Domino Jobs, reads their output, and syncs project files, and one searches Domino docs. Inside a Workspace there is nothing to configure. On a laptop, sign in once and the same tools are there. Keep the assistant and editor you already have.

Why it matters:

• Run analysis and training on Domino compute straight from your assistant, whether it sits in a Workspace or on your laptop
• Give every assistant 35+ Domino Skills, so it knows how Jobs, Apps, HPC, and the rest of the platform work before you ask
• Keep the audit trail intact, since every run an assistant starts lands in Domino like any other Job

Centralize your organization's AI access with LLM Gateway 2.0. Provide notebooks, apps, and coding assistants with a unified routing layer across OpenAI, Anthropic, Azure, AWS Bedrock, Google Vertex AI, and Domino-hosted models. Provider credentials stay secure at the app level, hidden from end users while automatically tracking usage against existing Domino identities. The Gateway also gives you a central control surface to switch between remote and locally hosted models based on data sensitivity.

Why it matters:

• Unify model access: Reach any approved external vendor or internally hosted model through one standard API tied to existing Domino identities.
• Control AI spend: Track usage automatically and enforce per-team or per-model budgets, including for coding assistants.
• Enforce security: Route sensitive workflows to private models and actively screen prompts and responses with built-in guardrails.

Domino Apps now include draft previews, so authors can test changes in a real App runtime before publishing, without touching the version users are on. Every app version is reproducible and can be restarted on demand. Approval gates sit in front of deployment, so Apps ship only after the sign-offs your policy requires.

Why it matters:

• Iterate draft previews, validating behavior before publishing, without disrupting production.
• Versioning is tied to reproducibility, so every app version is traceable and auditable.
• Govern app publication with approval workflows, so AI applications meet enterprise standards before they ship.

Submit and manage Slurm HPC Jobs directly from within the Domino, using existing commands and job scripts. It's real Slurm underneath, managed by Slinky, so nothing changes about how researchers work. Slurm handles scheduling and enforces the policies and limits admins set, while Domino handles monitoring, and governance, unifying HPC and AI/ML workloads under a single system of record.

Why it matters:

• Submit Slurm HPC Jobs from any Domino Workspace using native Slurm CLI commands
• Track job status, resource usage requested against used, and input and output files in one Jobs view
• Reproduce any past run, because every HPC Job lands in the Domino audit trail with full traceability