Cursor Lets Companies Self-Host AI Coding Agents for Enhanced Security

Cursor has announced the general availability of self-hosted cloud agents, a move designed to allow enterprises to run AI coding tools within their own infrastructure. The update addresses security and compliance concerns that have previously limited the adoption of autonomous coding agents in regulated industries. By enabling companies to keep source code and build data in-house, the AI code editor aims to expand its utility among Fortune 500 companies and organizations with strict data governance policies.

The coding platform, developed by Anysphere, was last valued at $29.3 billion following a $2.3 billion funding round in November 2025. The company stated that the new capability allows its agents to run code, tests, and development tasks locally while ensuring that sensitive artifacts never leave a company’s own environment. This shift comes as demand grows for AI tools that can operate within complex internal networks without exposing proprietary information to external services.

Security and Infrastructure Control

For AI coding agents to function effectively, they typically require access to private repositories, dependencies, and internal build pipelines. However, many organizations have been reluctant to grant external services this level of access due to legal and security restrictions. Cursor’s self-hosted solution reverses the standard cloud agent model by bringing the agents closer to where a company’s code and systems already reside.

Katia Baza, a Cursor engineer, explained the security implications in a blog post. She noted that self-hosted agents offer the benefits of cloud agents with tighter security control.

Self-hosted agents offer all the benefits of cloud agents with tighter security control: your codebase, tool execution, and build artifacts never leave your environment.

Katia Baza, Cursor Engineer

Under the new system, Cursor’s cloud agents run in isolated virtual machines that can be deployed on a company’s own infrastructure. These agents clone repositories, set up development environments, write and test code, and push changes for review. For teams with complex development environments, the self-hosted agents gain access to internal caches, dependencies, and network endpoints similar to a human engineer or service account.

Enterprise Adoption and Use Cases

Requests to self-host Cursor’s cloud agents had been surfacing among users for months prior to the official release. Developers discussed the need for such features in forum threads and community discussions, often citing the need to avoid exposing code externally or to connect more easily to internal systems. The update is now being utilized by companies including Brex, Money Forward, and Notion.

Graham Fuller, a Senior Software Engineer at Brex, highlighted the operational benefits of the self-hosted solution. He stated that the setup allows his team to delegate end-to-end software builds entirely to Cursor’s cloud agents while maintaining control over infrastructure access.

This self-hosted solution will allow us to delegate end-to-end software builds entirely to Cursor’s cloud agents.

Graham Fuller, Senior Software Engineer, Brex

Ben Kraft, a software engineer at Notion, described the update as a significant step toward making coding agents enterprise-ready. He noted that running agent workloads in their own cloud environment allows agents to access more tools securely and reduces the need for his team to maintain multiple stacks.

Self-hosted cloud agents are a meaningful step toward making coding agents enterprise ready.

Ben Kraft, Software Engineer, Notion

Technical Specifications and Limitations

Cursor says self-hosted cloud agents currently support up to 10 workers per user and 50 per team, with larger company-wide deployments available on request. Self-hosted agents can be enabled through the Cursor dashboard once a worker is set up in the target environment. The company’s documentation indicates users can spin up a worker on a laptop, a devbox, or a remote VM to utilize local dependencies and network access.

Despite the local execution capabilities, self-hosting does not remove every challenge. Companies still need to deploy and manage the infrastructure these agents run on. The agents’ planning and coordination still take place in Cursor’s cloud, with only execution handled locally. This hybrid model ensures that while code remains secure, the orchestration logic remains centralized.

Market Context and Competition

Cursor’s move toward self-hosted agents sits within a broader effort to make its tools usable in environments that have historically been difficult to reach. The company says its broader platform is currently used by more than two-thirds of the Fortune 500. This expansion comes as competition intensifies from standalone coding tools and major model providers, including OpenAI, Anthropic, and Google, which are building their own agent systems.

The company has been expanding its agentic capabilities, including always-on agents that can handle code reviews and bug triage. It has also open-sourced security-focused agent templates. These moves coincide with the release of Composer 2, a model designed to handle longer coding tasks at lower cost. However, the model has drawn scrutiny after Cursor acknowledged it is built on top of Moonshot AI’s Kimi K2.5, a Chinese open-weights model that was not disclosed at launch.

By removing barriers related to data sovereignty and network security, Cursor aims to differentiate its product in a crowded market. The self-hosted option expands where the product can be deployed, potentially unlocking adoption among teams with complex infrastructure and security requirements that previously prevented the use of external AI coding assistants.