Compute
- Compute pools
- Organize your cloud hardware into pools and assign them to the teams that need them. The platform schedules workflows, workstations, and endpoints onto your nodes so your teams can focus on their work rather than infrastructure.
- Multi-cloud and external compute
- Extend the platform by attaching additional cloud providers, GPU providers, or on-premises clusters alongside your primary cloud.
- Autoscaling
- Your workloads and deployments scale up automatically under load and back down when demand drops, so you pay for what you use.
- Cloud workstations
- Give your data scientists browser-based or VS Code-connected development environments backed by your compute pools, including GPU instances.
Workflows
- Metaflow-based
- Your teams write workflows in Python using Metaflow, with decorators to declare compute requirements, dependencies, and scheduling in their code.
- Automatic environment builds
- Declare dependencies in code and the platform resolves packages from your governed channels, builds container images, and caches them for subsequent runs.
- Model-serving endpoints
- Deploy long-running services like APIs, dashboards, or LLM inference endpoints with autoscaling and managed lifecycle.
Package governance
- Curated package channels
- Give your teams access to Anaconda’s security-reviewed package repository, scoped to the teams and projects that need them.
- CVE-based channel policies
- Filter packages by vulnerability score and status before they ever reach a workload. Configure policies per channel to match your organization’s security posture.
- Private and external channels
- Register your organization’s internal channels or third-party sources (such as conda-forge) alongside Anaconda channels.
Model governance
- Anaconda model catalog
- Provide your teams with a curated set of security-reviewed open-source LLMs, each with rich metadata including license, publisher, parameter count, and an AI Bill of Materials (AIBOM).
- Model access policies
- Control which models each team can browse and download based on conditions like publisher, license, size, or country of origin.
- Server-side enforcement
- Blocked models are never visible to users and cannot be downloaded regardless of how the request is made.
Access control
- Perimeters
- Separate your teams and projects into isolated boundaries, each with its own channels, policies, compute access, and user privileges.
- Role-based access
- Assign Admin or Member roles at the platform level, and grant View or Execute privileges on individual perimeters.
- Machine users
- Set up dedicated identities for CI/CD pipelines and automation, scoped and audited separately from your human users.
- SSO integration
- Connect your existing identity provider so your teams use the credentials they already have.
Monitoring
- Audit logs
- Track administrative actions across the platform, including user management, policy changes, and workstation lifecycle events.
- Perimeter event logs
- Review workflow execution and trigger activity within each perimeter (72-hour retention).
- Compute utilization and cost visibility
- See how your resources are being consumed across teams and workloads.