Building AI workflows at scale means solving three problems at once: finding the right model, working confidently in your environment, and maintaining a clear record of every decision along the way.

Today we’re announcing enhancements across AI Catalyst and the Anaconda Platform that address all three. Smarter model discovery with quantifiable security evaluation, a new model context protocol (MCP) server that makes your AI coding assistant genuinely conda-aware, and a full audit trail for package decisions. These capabilities will enable your team to move from experimentation to production with confidence.

AI Catalyst: Smarter Model Discovery for Production AI

Selecting the right AI model for production shouldn’t require weeks of evaluation. Our latest AI Catalyst enhancements introduce intelligent filtering and expanded benchmarking that dramatically reduce model selection overhead while ensuring compliance and security from day one.

Enhanced Model Discovery

New tagging and filtering capabilities let teams discover models matching exact requirements:

  • Responsible AI scoring surfaces quantifiable security metrics including risk scores, OWASP Top 10 for LLMs vulnerabilities, and MITRE ATLAS mappings
  • Tool-calling capability tags identify models optimized for agentic frameworks and API integrations

Expanded Benchmarking

Evaluate models on what actually matters for your work. New benchmarks for data science code generation (DS-1000), instruction-following precision (IFBench), scientific computing (SciCode), and Python code quality (HumanEval/LiveCodeBench) ensure you’re selecting models based on enterprise-relevant performance, not just generic metrics.

AI Catalyst's Responsible AI surfaces quantifiable security metrics, including overall risk scores.

NVIDIA Nemotron Models with Integrated Governance

AI Catalyst now includes NVIDIA’s Nemotron foundation models with the same vulnerability scanning, compliance documentation, and reproducibility controls you already have for Python packages, eliminating governance gaps and workflow context-switching. Learn more about Anaconda’s NVIDIA partnership.

NVIDIA Nemotron models with full quantization are now available in AI Catalyst.

MCP Server: An AI Assistant That Knows Your Environment

AI coding assistants are great at writing Python, but until now, they had no real understanding of your conda environments or the broader ecosystem. They guess at package names, miss version conflicts, and generate code that breaks the moment it hits a real environment.

Anaconda MCP is a new model context protocol server that gives AI coding tools direct access to your conda environments. Your AI assistant can now see what’s installed, create new environments to spec, and reason about dependencies against your real system state. One install, one configuration and your AI tools are conda-aware, with platform teams in control of exactly which capabilities are exposed.

This ships alongside conda-meta-mcp, built by Daniel Bast and Jannis Leidel from the conda open source community. Where Anaconda MCP covers your local environment, conda-meta-mcp covers the broader ecosystem: authoritative, real-time package metadata across conda-forge and other channels. This is the first chapter of Anaconda’s broader MCP story, with more to come later this year.

Quickly leverage Claude and MCP Server to get a glimpse at your environments and packages.

Audit Trail for Package Decisions

Every organization managing packages at scale needs to know what changed, when, and why. The new audit trail gives you a timestamped, searchable record of every package addition and removal, including exactly why it happened. Events are captured automatically across multiple event types, triggered either by mirror runs or manual actions. Examples include:

  • Mirror syncs: logged when packages are added as channels sync from upstream sources like conda-forge
  • Manual uploads: captured with user and timestamp whenever a package is added outside of a mirror sync
  • Policy-driven removals: recorded when packages are automatically removed for failing channel policy rules, including CVE-based filtering

Filter history by package, date, channel, source, or reason code. You can easily integrate audit data into existing compliance workflows, and bulk export means reviews no longer require manual digging. Every user can fully understand how and why their channel content changes over time.

Audit trail gives you a timestamped, searchable record of every package addition and removal.

From Experimentation to Production, Faster

These enhancements represent a fundamental shift in how enterprises approach AI development. By combining intelligent model discovery, environment-aware AI tooling, and transparent package governance, Anaconda gives organizations everything they need to move from experimentation to production without sacrificing speed or trust.

  • Reduce model selection from weeks to hours. Intelligent filtering automatically excludes models that fail technical, security, or compliance requirements before evaluation begins.
  • Deploy with quantifiable security confidence. Models are evaluated against OWASP and MITRE frameworks with robustness scores you can present to security and compliance teams.
  • Keep development teams in one governed ecosystem. Code, dependencies, infrastructure, and models are all under consistent security and compliance controls with no context-switching.
  • Satisfy compliance reviews without manual work. Access a full audit record of every package change, exportable in bulk, with API access for integration into existing workflows.

Get Started

Learn more about AI Catalyst and the Anaconda Platform, or contact our team to discuss how these capabilities can accelerate your organization’s AI development.