Cursor cloud agents now ship full web apps: Postgres, auth, hosting

The prompt-to-deploy loop moves inside the editor — alongside realtime voice agents and Buoyant's open-source KV-cache router for self-hosted vLLM.

Nowline NOV 9 10:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • Prompt to a live app, without leaving Cursor

    Cursor's new Build Live Apps lets cloud agents generate a web app, provision a Postgres database, wire up Cursor SSO auth, and deploy hosting — with security checks (exposed secrets and the like) that must pass before anything goes live. It's on every paid plan with cloud agents, deploying from Cursor Origin or GitHub (GitLab "coming soon"). A weekend build can now go prompt-to-shipped, authenticated and hosted, inside the editor.

  • Call your coding agent, literally

    Cursor added realtime voice chat with agents across the Agents Window, IDE, web, and iOS — say "Hey Siri, call Cursor," then mute or hang up from the Lock Screen. It runs on Grok Voice with selectable voices and zero data retention under Privacy Mode, and gives spoken progress updates on running tasks and diffs. Useful for steering a long agent run hands-free.

  • An open, cache-aware router for vLLM fleets

    Buoyant — the team behind Linkerd — open-sourced Kinference (Apache 2.0), an inference control plane for Kubernetes and vLLM. It watches vLLM KV-cache events via NVIDIA's dynamo-kv-router and load-balances OpenAI-format requests across endpoints, with a kinfctl CLI to validate routing policies first. It's alpha — no Anthropic-style /v1/messages, rate limiting, or multi-replica support yet — but it's a self-hostable way to route by what's already cached.