You ask your coding agent to fix a bug. A few minutes later, it has read files, installed a dependency, run commands, and called a tool through MCP. That is a lot of access behind one request.
As agents take on more of the work, builders need a clear say in what they can do. You should be able to choose your model, understand what you are connecting, and set boundaries without spending your day assembling security tooling.
That is why Anaconda, the company behind Kilo, has joined the Open Secure AI Alliance, as we announced today.
What we are building together
Founded by NVIDIA and led by the Linux Foundation, the Open Secure AI Alliance brings together technology companies, researchers, and government partners to develop and share open technologies for securing AI. Members include Anaconda partners NVIDIA, Microsoft, and Amazon, alongside Cloudflare, Red Hat, and others across the AI and security ecosystem.
The intent is practical: give builders and defenders tools they can inspect, adapt, and run in environments they control. That includes open models, security testing methods, and protections for the software and tools agents depend on.
Anaconda plans to collaborate with fellow members on three priorities:
Test the whole agent workflow. Find failures that unfold across multiple decisions and tool calls, including when a retrieved document changes an agent’s plan and triggers an unintended action.
Secure agent connections. Improve how teams scan, vet, and control access to tools and MCP servers.
Show that protections work. Record agent decisions and report incidents so teams can verify which security controls ran and whether they held.
For builders, the goal is more consistent protection across the models, tools, and infrastructure you choose. Shared approaches can reduce the security work each team has to figure out independently and help improvements reach more of the ecosystem.
Open models put more choices in your hands
Kilo is open source and model agnostic. Use a frontier model for a difficult refactor, an efficient open model for routine work, or a local model when the project calls for it. Bring your own API keys if that works better for you.
NVIDIA Nemotron models are part of that choice. They give Kilo builders more options for coding, reasoning, and working across code, data, and images, with different tradeoffs in capability, performance, and cost.
Open weights also give teams options to customize models and run them on infrastructure they control. For a sensitive codebase or a project with specific data requirements, where a model runs can matter as much as how well it codes.
An open ecosystem also gives more people the ability to inspect the tools, find problems, and improve the defenses everyone depends on.
Make safer choices part of the workflow
Secure by default has to mean something practical when you sit down to build: trusted components, clear boundaries for your coding agent, and protection for the applications you ship. Kilo brings that approach into your workspace, backed by the broader Anaconda platform.
In Kilo Desktop, you can create and manage conda environments with vetted packages from Anaconda. Through AI Artifacts, you get packages and curated models with provenance and vulnerability information, giving you a trusted supply chain from the start.
As your agent works, tool permissions let you allow actions, require approval, or block them. You can automate routine work while keeping consequential commands and changes subject to review. Those boundaries matter when you connect MCP servers that give agents access to additional tools and data.
That protection extends to the AI applications and agents you build. The Anaconda platform’s security capabilities, powered by Enkrypt AI, help you test how agents can be manipulated before release, enforce policies on their actions in production, and retain evidence for continuous compliance.
From the packages you start with to the actions your agents take, the goal is to make secure development part of how you build every day.
Keep building with open choice
Kilo gives you the freedom to choose how you build. Our work with Anaconda and the alliance supports that freedom by helping strengthen the foundation beneath it. When you use the right AI stack, security and model choice can go hand-in-hand.
We want you to get more done with models you choose, controls you understand, and protections that keep pace with what your agents can do.
Try all of the NVIDIA Nemotron models in Kilo, review your agent permissions, and keep building with Kilo by Anaconda.
