Anaconda Platform Adds Governed Packages and Models to Every Flow Step

Approved packages through one decorator

Flows in Anaconda Platform are Python classes in which each method is a step. They are built on Metaflow, the open-source framework that came to Anaconda with the Outerbounds acquisition in April 2026. To use Anaconda's trusted packages, a developer adds the @anaconda_base decorator to a flow and lists the Python version and the packages needed. In Anaconda's example, a flow declares Python 3.12 along with pytorch 2.6.* and pandas 2.2.*.

According to Anaconda, the platform then builds OCI-compatible images from its governed channels, so no environment has to be set up by hand. Each step draws on curated, security-reviewed channels, has the organization's package policies applied before anything installs, and uses packages backed by security metadata and software bills of materials (SBOMs) that document components, licenses and provenance. The channels draw on Anaconda's Trusted Distribution, which the company describes as 19,000+ vetted packages built, tested and digitally signed by Anaconda engineers.

The platform resolves dependencies against a perimeter's channels and bakes the result into a container image, so the same environment runs locally or in production. A flow can be run locally with --environment=anaconda or on a cluster with --environment=fast-bakery.

Policies set by administrators, inherited by flows

Administrators set package policies on each perimeter's channels. A policy is a set of exclusion rules based on Anaconda's package security metadata, such as common vulnerabilities and exposures (CVE) score, status or license family. Each channel offers a recommended default policy that can be customized. Saving a policy rebuilds the channel's package index without the excluded packages, so they are unavailable to any flow in the perimeter and developers change no code.

Because each perimeter has its own channels, a production environment can run a stricter policy than development. Moving a flow to production goes through the team's existing CI/CD pipeline, including approval steps such as code review, test runs or sign-offs. The code stays the same, and the production perimeter's policies apply from that point on.

A model catalog filtered by policy

Steps that need a model pull it from Anaconda's model catalog, which holds 77+ curated models. According to the company, each is automatically sourced, quantized and benchmarked, then validated by Anaconda's team before publication. Models come with an AI Bill of Materials (AIBOM), license documentation and benchmark results.

Administrators write rules from model metadata: publisher, license, country of origin, model size and expected memory requirements. One example rule is that production only uses models whose license allows commercial use. Approved models can be used inside a flow step with the @anaconda_models decorator, downloaded for local work, or run as a dedicated inference deployment with its own API endpoint.

• Each model is available at more than one quantization level, trading some precision for a smaller footprint and faster response.
• Models can be deployed to a GPU-autoscaling endpoint in the cloud or loaded locally with Desktop.
• Both paths expose the same API, so calling code does not change between testing and production.
• Some models include a built-in chat interface.

Responding to security advisories and what it means

Each perimeter's channels list their packages and the known CVEs for each. Anaconda uses CVE-2025-32434 as an example: it affected how PyTorch loads model files in versions up to 2.5.1 and was fixed in 2.6.0. Since flows declare packages in code, finding affected flows is a search of repositories, and the fix is a one-line change to the pin followed by a rerun. The platform keeps artifacts from every run, so results from before and after the change can be compared using Metaflow's Run object.

For teams, the pitch is that developers declare what a flow needs, administrators set policies once, and the platform resolves each step against those rules. Anaconda says existing flows keep working as they are. The company offers demos in a customer's own cloud account and has published documentation.