Anaconda Acquires Kilo Code to Govern 10 Trillion Tokens a Month

Anaconda has acquired Kilo Code, the open-source, model-agnostic AI coding agent used by more than 3 million developers. The deal is a direct bet that the next big enterprise AI problem is not capability, but governance: who controls the models, the spend, and the data flowing through them.

Zero to 3 million in 16 months

Anaconda, a company that provides governed, open-source packages and environments for enterprises, has acquired popular open-source coding agent Kilo. Kilo grew its agentic engineering platform from zero to a thriving open-source community of 3 million developers in only 16 months. That kind of organic growth, driven entirely by developer word-of-mouth, is what made Kilo an attractive target.

Kilo itself is a fork of Roo Code, an open source AI coding agent for VS Code. When Roo Code sunset its IDE extension in favor of a cloud-only agent, it sent a wave of its user base looking for alternatives, with Kilo positioning itself as a landing spot. The project now has more than 26,000 GitHub stars and 3,000 forks, and Kilo alone orchestrates almost 10 trillion tokens per month across more than 3 million developers.

The people behind the deal

The key executives are:

• David DeSanto, CEO of Anaconda, who joined in October 2025 after six years as GitLab's chief product officer
• Scott Breitenother, CEO and co-founder of Kilo Code, who will continue leading the Kilo team post-acquisition
• Sid Sijbrandij, Kilo co-founder and GitLab co-founder and executive chair, who championed the deal

Kilo's numbers go some way toward explaining why Anaconda wanted it: more than 3 million developers routing close to 10 trillion tokens a month through the platform, across 500-plus models from over 60 providers. That scale, combined with Anaconda's existing enterprise reach, is the core logic of the acquisition.

The enterprise AI governance crisis

The strategic rationale is rooted in a problem that is getting worse fast. Enterprise AI spend is growing faster than anyone's ability to account for the spend. AI builders are being told to move fast, and the token consumption accumulates invisibly, spread across dozens of tools, work accounts, and personal accounts. It leaves no single view of where it all goes.

Anaconda CEO David DeSanto frames it as a false binary that most enterprises are stuck inside: either lock everything down to one tool and one model provider, or look the other way while developers use whatever they want with zero visibility. 80% of AI projects fail to make it to production. The problem isn't whether or not to leverage AI, the hard problem is doing that safely, and affordably, at scale.

There is also a dependency risk that gets less attention. Many enterprise development workflows today are entirely dependent on a single model provider. If that provider has an outage, changes their pricing, or retires a model, work stops. No enterprise would accept that kind of single point of failure in any other critical system.

What Kilo actually does

Kilo is an agentic coding assistant, meaning it does not just autocomplete lines of code. It can plan, execute, and iterate on multi-step engineering tasks autonomously. Its core differentiator is model neutrality: rather than tying developers to one AI lab, Kilo lets them route work across whichever provider they choose. Key capabilities include:

• A model gateway that routes intelligently across 500+ models from 60+ providers, including open-weight and frontier options
• Multi-agent orchestration, where several agents collaborate on a single task, each matched to the best model for that subtask
• Works inside VS Code, JetBrains, and the CLI, so developers do not change their environment
• Analytics giving organizations a consolidated view of AI development activity and spend across the enterprise
• Self-hosted and air-gapped deployment options for sensitive workloads

Tools like OpenCode, Cline, and Aider have found traction by sitting a layer above the models themselves, giving developers a way to dodge unpredictable token bills and vendor lock-in. Kilo competes in this same category, but with the added weight of Anaconda's enterprise distribution behind it.

A platform built by acquisition

This is Anaconda's second major acquisition in rapid succession. In June, Cursor acquired Continue, another open source, model-agnostic assistant, folding one of the category's earlier standard-bearers into a closed, commercial IDE. Anaconda is taking the opposite approach: keeping Kilo open and folding it into a governed platform rather than closing it off.

Earlier this year, Anaconda also acquired Outerbounds, the company behind Metaflow, Netflix's open-source ML orchestration framework. Where Kilo handles the moment code gets written, Outerbounds governs what happens once that code needs to run reliably in production. Together, they give Anaconda coverage of the full AI-native development lifecycle, from first prompt to production deployment.

Anaconda itself is no small player. Anaconda provides governed, open-source packages and environments for enterprises, with over 52 million users and 95% of the Fortune 500 relying on the platform. It raised a $150 million Series C in 2025, valuing the company at around $1.5 billion.

Who wins, who watches carefully

The clearest winners are enterprise teams that have been caught between developer freedom and IT governance. With Kilo, Anaconda provides a secure platform where agents can get work done at enterprise scale, enabling security and governance without the need to trade off velocity.

Individual Kilo users are also protected in the near term. Kilo remains available today with no changes to existing products, plans, or support for current users. Anaconda takes over stewardship of Kilo's GitHub organization and developer community, and has committed to keeping the project open source.

The companies most exposed are the ones building closed, single-provider coding tools. Cursor's acquisition of Continue illustrates how contested this layer of the stack has become. Anaconda is now a well-capitalized, enterprise-credentialed competitor in that same fight, with an open-source community already attached.

The token efficiency angle

Beyond governance, there is a concrete cost argument. Early results from customers running AI agents through the Anaconda Platform are reporting 30 to 50% reductions in token consumption. The mechanism is intelligent routing: rather than burning tokens through repeated correction cycles, the Anaconda MCP server ensures agents get the right context on the first pass. At 10 trillion tokens a month, even a modest efficiency gain translates to significant savings.

What comes next

The integration roadmap is phased. Kilo is available today inside VS Code, JetBrains, and the CLI with no changes. Over the coming months, Anaconda plans to connect Kilo to its governed package catalog, vetted model registry, and orchestration layer, so that a project can move from a developer's first prompt to a production workload without switching platforms or losing policy enforcement along the way.

Collectively, this illustrates how contested this layer of the stack has become. And Anaconda is jostling for its seat at the table. With 3 million developers already using Kilo and 52 million on the broader Anaconda platform, the combined community gives Anaconda a meaningful distribution advantage as this market consolidates. The deal terms were not disclosed.