Hey HN,

If you’re building local AI agents (or using tools like Cursor/Claude Code), giving them a place to safely execute code and manipulate files is a headache. Running untrusted LLM-generated code directly on your machine or local Docker is risky, and managing remote SSH keys for an automated agent is clunky.

I built Oblien to solve this. It provides instant, hardware-isolated cloud VMs, but the actual interesting part is how you interact with them: the Runtime API.

Every workspace runs an internal HTTP/WebSocket server on port 9990. Instead of establishing an SSH connection, your local agent just makes standard API calls to control the remote machine securely.

It exposes everything an agent needs to do real work:

• Command Execution: Run POST /exec to execute bash commands. It supports synchronous runs, or you can stream stdout/stderr in real-time via SSE for long-running tasks. • Files & Search: A full file API to list, read, and write files. It also has a native /search endpoint powered by ripgrep, so agents can find code without pulling the whole repo locally. • WebSockets & Terminals: You can open interactive PTY sessions multiplexed over a single WebSocket connection. It even supports streaming filesystem events (create/write/delete) to watch directories.

How local agents connect securely: The workspaces are network-dark by default—outbound internet is on, but inbound is completely blocked.

To let your local agent in, you hit the Oblien API to get a short-lived Gateway JWT. The agent uses this token as a Bearer auth header against workspace.oblien.com. Our gateway decodes the routing info from the JWT and securely tunnels the HTTP request through the firewall directly to the VM's internal server.

It makes orchestrating remote compute feel like using a standard web API.

For example, to run a command inside the remote VM, your agent just does this: curl -X POST "https://workspace.oblien.com/exec" \ -H "Authorization: Bearer $GATEWAY_JWT" \ -d '{"cmd": ["npm", "install"]}' Docs and API reference: https://oblien.com/docs

Would love to hear what you think of the API design, or if there are specific endpoints you'd want added for your own agent workflows.

also You can use it like Vercel—spin up a VM, add workload, expose a port, and attach a custom domain for instant preview URLs. But unlike serverless platforms, you aren't restricted. Because every workspace has an injected Runtime API (port 9990), you can programmatically execute commands, edit files, and multiplex terminals over WebSockets right from your local machine or CI pipeline. You get the hosting of a PaaS, but the low-level control of a raw Linux VM