Artificial intelligence development firm Anaconda Inc. is enhancing its enterprise AI platform with tools to coordinate groups of AI agents, test their security and move applications into production.
Today’s introductions continue the company’s expansion beyond its roots in Python software distribution into broader AI development and management. The announcement draws on three recent acquisitions: AI coding company Kilo Code Inc., security specialist Enkrypt AI Inc. and orchestration provider Outerbounds, the business name of Step Computing Inc.
“We are moving Anaconda from being known as a Python package company into being known as an AI-native development platform, essentially helping people build their own enterprise AI applications,” said Chief Executive David DeSanto.
DeSanto said the expansion addresses enterprise concerns about coordinating autonomous work, controlling costs and understanding what agents do when they access software and data. Those concerns were amplified by this summer’s much-publicized incident in which OpenAI agents escaped the company’s test environment and hacked into systems at AI model repository Hugging Face Inc. without informing their human supervisors.
Such “agent swarms” are central to Anaconda’s new release. In a swarm, a task agent delegates parts of a project to subagents that can work simultaneously, exchange information and select different models for different jobs. It enables agents to tackle more complex problems and make decisions without requiring human oversight. Anaconda provides a secure message board through which agents can communicate and users can observe their interactions.
For example, DeSanto said a swarm working on a Jupyter notebook inside Anaconda recently identified an incorrect configuration, corrected it and notified the other agents that their original plan needed to change.
The company is bringing swarms into Microsoft Corp.’s Visual Studio Code through technology acquired with Kilo. The new Kilo Desktop combines software development, data science and Python environment management, including notebook sessions where people and agents can edit and execute cells together. The workspace provides access to more than 500 models, supports local model execution and includes automatic model routing to optimize spending.
Anaconda said its own research found that 63% of respondents are moving toward swarms in some form. Some have already deployed hundreds of agents, including groups focused on development and security testing, DeSanto said.
Managing groups of agents working on complex, multifaceted tasks requires more visibility than individual agents typically provide, DeSanto said. Logs and the shared message board in the Anaconda platform let customers reconstruct interactions and inspect activities such as file access or configuration changes. Logs can run into thousands of pages, and Anaconda is working to make large volumes of activity easier to digest.
Agentic red teams
Security capabilities acquired with Enkrypt include autonomous red-teaming agents that challenge models, agents and Model Context Protocol connections across more than 300 attack categories. MCP allows agents to connect to outside tools and information.
The tests inform guardrail development and are checked in production as applications change. Runtime controls can approve, modify or block risky behavior across agents, tools and retrieval-augmented generation systems.
“You can use the red teaming data in development to write better guardrails for what you’re trying to do,” DeSanto said. “And then you can revalidate that in production with the same red teaming.”
To minimize the risk of security breaches, Anaconda provisions agents initially with no access, a technique called least privilege. The platform also features a public Agent Incident Registry with source-backed records of reported agent incidents.
Agentic security is a growing concern. The company said recent Enkrypt research found vulnerabilities in 73% of the agent tools examined across more than 25,000 MCP servers. That doesn’t mean every gap was a lurking disaster; DeSanto said findings include potential data leakage and other exposures.
Simplified workflows
Outerbounds supplies orchestration technology for repeatable workflows and reproducible environments. The platform carries governed packages, models and security and licensing policies from development into production. Its FastBakery technology builds container images incorporating the open-source conda manager and dependencies documented in the PyPI package repository, including native libraries.
Anaconda’s curated catalog now contains 77 models, many of which are freely modified open-weight models. DeSanto said demand for such models is growing, particularly among regulated companies that need control over where inference runs and how data is handled. Models undergo performance and security testing before entering the catalog.
Anaconda says 95% of the Fortune 500 use its software. While that figure includes both free open-source and paid use, DeSando said he believes most large firms have at least some paid components. Adoption typically begins with a developer or team, then expands as companies seek standardized environments, governance and supply chain visibility, he said.
Integration of the acquired companies’ technologies is ongoing. Outerbounds is fully integrated into the Anaconda platform, and Enkrypt security scores appear on model cards, but additional guardrails, combined model catalogs and enterprise controls remain on the roadmap.
“Early next year, you should have a fully integrated package you can purchase from Anaconda,” DeSanto said.
The enhanced platform carries a separate license from Anaconda’s legacy tools, with a platform fee plus consumption charges. Kilo and Enkrypt pricing will remain unchanged until integration is complete. Anaconda publishes many of the pricing details on its website. The company is retiring its legacy workbench and package security products and migrating cloud package security customers to the new platform.