jurniti
Always-on AI agents in Firecracker microVMs
49 followers
Always-on AI agents in Firecracker microVMs
49 followers
jurniti is a private always-on box for your AI agent. Each agent gets its own Firecracker microVM with a KVM hardware boundary, the same primitive AWS uses for Lambda, not a shared Docker container. Bring your own model keys. They stay inside the guest. Pay, and the box is live in about three minutes: in-browser terminal, persistent disk, your subdomain. Claude Code, Codex, Hermes, OpenClaw, and other upstream agents. From $25/mo. 30-day money-back on the first purchase. No free trial.
the microVM/KVM boundary and BYOK-stays-in-guest framing is a genuinely stronger isolation story than most "agent sandbox" products that are just a shared container with an env var. since these are always-on rather than spun up per-task, what's the kill-switch story if an agent inside the box starts making runaway API calls on your model key - is there usage-based alerting or a hard cap, or is it on the user to notice the bill?
@galdayan Honest: the token bill is yours.
We never see the key, so we cannot cap spend. The kill switch on our side is Stop on the VM. Egress off the box is rate limited.
Token alerts live at OpenRouter or whoever issued the key.
appreciate the direct answer, especially "we never see the key, so we cannot cap spend" being stated plainly instead of hedged. rate-limited egress plus alerts at the token issuer's end is a reasonable belt-and-suspenders setup, though it does mean the actual dollar cap still lives outside jurniti itself. makes sense given the whole pitch is the key never leaving the guest.