News · Cursor · 2 min read
Event date: · Published:
Cursor lets cloud agents run their tool calls on self-hosted machines
On September 2, 2026, Cursor added self-hosted machines so that cloud agents could run their tool calls on infrastructure the customer controls. Code, build artifacts and credentials stayed inside the customer's network.
Archive entry: this covers an event from September 2, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.
On September 2, 2026, Cursor announced self-hosted machines, which let cloud agents run their tool calls on infrastructure the customer controls. Cursor explained the feature in its blog post on running cloud agents on machines you manage.
What changed
- Tool execution stayed inside the customer's network: code, build artifacts and credentials remained on the customer's infrastructure while the agent's tool calls ran locally.
- My Machines linked one laptop or VM for personal use.
- Team pools were named worker queues that scaled with demand, could pause idle machines and brought them back during a reconnect grace period.
- Cloud agents could run on infrastructure such as AWS Lambda, Coder, Cloudflare, Daytona, Modal, Namespace, Vercel and E2B.
- Linux and Mac self-hosted workers could operate a desktop, including clicks, typing, screenshots and browser control.
What it means for people building with AI
Companies that could not send code or credentials to a vendor cloud could still use Cursor's cloud agents, because the tool calls ran on machines they managed. For a single developer, that meant linking one laptop or VM. For a team, it meant a named pool of workers that scaled with demand.
Sources and dates
Both the blog post and the Self-hosted machines changelog entry carry the date September 2, 2026.
Sources
- Run cloud agents on machines you manage (Cursor, September 2, 2026, official)
- Self-hosted machines (Cursor, September 2, 2026, official)
Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.