Most AI agents do their work inside a vendor's sandbox — your code copied up, their model bundled in, their off switch. The Kynver runtime turns that around: the agent runs on a box you own, with the model you choose, and you can stop it any time.
Where an agent runs decides who holds the keys. These aren't features on a roadmap — this is how the runtime works today.
The agent runs on hardware you control — a desktop that stays on, a spare laptop, a server in your office. You bring it online with one command, and you can stop it at any moment. Nothing runs that you didn't choose to run.
The work happens where your repositories and logins already live. Your code isn't copied into a vendor sandbox to be worked on — the agent checks out branches, runs tests, and opens pull requests from your own machine, using access you already have and can revoke.
Run the work through the coding-agent subscription you already pay for, or point the runtime at any compatible model endpoint — including one running entirely on your own hardware. Switch anytime; your agent's memory lives with you, not inside any one model.
Before your machine acts on an instruction, it cryptographically checks that the instruction was really issued by your workspace. Anything that doesn't check out is refused — your box takes orders from you, and it can prove it.
The work leaves a verifiable record. Access to your agent's memory is written into signed, tamper-evident receipts you can export and check yourself, and the work itself ships as pull requests you review before anything lands.
Once your workspace is approved for early access, standing up a box takes three commands. The setup guide walks through each one.
One package from npm: npm install -g @kynver-app/runtime. Needs Node.js 24.11 or newer on a machine that can stay on.
Run kynver bootstrap. You approve the machine from your browser — no keys to paste — and it's linked to your workspace with its own scoped credential you can revoke at any time.
Run kynver start. Your agent comes online and picks up work; you watch it live in the Command Center. Stopping it is as simple as pressing Ctrl+C.
Hosted agent sandboxes are convenient — and everything about them belongs to someone else. Running the agent on your own machine changes who's in charge of every part of the arrangement.
The runtime is in gated early access while we harden it on real workloads. Request access with a free Kynver account — every request is reviewed, and approved workspaces get the full runtime.