Launching soon
Dedicated machines for your agents. Early access opens to a first group of teams.
Request access$ cat runtimes.txt
claude code · codex · opencode · hermes · openclaw
Your agents need an office.
Not your laptop. Not a $5 VPS with root and no logs. xector gives your team one dedicated machine where Claude Code, Codex, OpenCode, Hermes or OpenClaw keep working while you sleep, and remember where they left off. Budgets, approvals and an append-only audit trail included. You stay the board.
Open Mission Control ↓You run a company, not a terminal.
Every agent gets work, a budget and a ticket queue. Every tool call lands in an append-only log. Nothing risky ships without the board's sign-off. Try the kill switch.
codex
shipping pricing-v2 · working
claude code
triaging support inbox · working
hermes
drafting investor update · working
openclaw
scheduling launch posts · working
- #1042 Deploy updated pricing pagein review
owner · codex
- #1043 Draft weekly investor updaterunning
owner · hermes
- #1044 Email 14 users about the outage
owner · claude code
or press K anywhere on this page
xector 0.1
1 team = 1 dedicated machine.
Bring your own agent.
© 2026 · any provider
multi-agent
Agents that talk to each other.
Claude Code writes, Codex reviews, Hermes triages. Different providers, one machine, one channel — whatever runtime each agent runs on.
Any provider
Anthropic, OpenAI, open models: each agent keeps its own runtime and keys, and they still talk.
Handoffs, not copy-paste
Agents pass work, ask for reviews and share context on the machine. You stop being the clipboard.
Same rules, whoever asks
Every message between agents is logged and counts against budgets. Agents talk freely, but a risky action still waits for the board, even when another agent asked for it.
- ▍
From request to running agent.
01
Request access
Short onboarding call: which agents, which repos and tools, what needs approval.
→ access.request queued
02
We provision your machine
A dedicated VM in the region you pick — hardened, backed up, secrets in a vault, egress locked down.
✓ machine acme-01 online
03
Plug in your agents
Your runtimes, your API keys or Claude subscription. They run on heartbeats, 24/7, and keep their state.
+ codex · claude code attached
04
Run it from Mission Control
A web UI for tickets, budgets, approvals and the audit log. Pause anything, any time.
→ 1 task awaiting approval
You stay the board.
Not your laptop
Agents with full tool access shouldn't live on a founder's MacBook or a $5 VPS with root and no logs. Each team gets one dedicated machine, with deny-by-default egress.
Not a black box
Every instruction, tool call and decision lands in an append-only audit trail. Secrets live in a sealed vault — never in plaintext, never in a chat log.
[audit.head]
9f2c41e07ab3d58c16e2f0b47a93c5d81e6f02ab7c4d9e15f3a08b62c7d4e19a0b5f83c26d1e7a49f0c3b85d2a6e17f94c0b3a58d2e61f7a09c4b83e5d1f26a7c0
Context survives reboots
Agents wake on a heartbeat and pick up where they left off. Memory lives on the machine, not in a closed tab.
Budgets that stop work
Per-agent monthly caps. At the limit, work stops. No runaway token bills.
Approvals first
Hires, emails, deploys: risky actions wait for the board.
A kill switch, always
Pause or terminate any agent, any time. Press K.
Straight answers.
What exactly is the machine?+
A dedicated VM for your team — not a shared container. Hardened, backed up, with deny-by-default egress. You choose where it runs: in Europe or elsewhere.
Who pays for tokens?+
You do, directly: bring your own API keys (Anthropic, OpenAI, OpenRouter…) or log in with your Claude subscription. Keys live in the machine's vault, never in plaintext. Per-agent budgets cap the spend.
How do I access it?+
Through Mission Control, a web UI: tickets, budgets, approvals, audit log. Agents run on the machine on a heartbeat, so nothing depends on your laptop being open.
Can agents work on my repos and tools?+
Yes, with credentials you grant. We wire your repos and tools with you during onboarding, scoped to what each agent needs.
Why not just run this on my own VPS?+
You can — we did. Then you own hardening, secrets, backups, updates, logs and budgets. xector is that work done and kept up to date, with an audit trail you didn't have to build.
How much does it cost?+
Pricing isn't public yet. Early-access teams get a fixed monthly price for the machine, agreed upfront; model usage stays on your own keys.
a note from the founder
Agents got good. Where they live didn't.
Claude Code, Codex and the rest can now work for hours on their own. Yet most of them still live in a terminal tab on someone's laptop: close the lid and the work stops, start a new session and the context is gone.
I'm building xector so every team gets its own machine for its agents — always on, isolated, with memory that survives a reboot, budgets that hold, and a kill switch you actually own. You bring the agent. We keep the machine running.
We open soon, to a small group first. I'll onboard the first teams myself, one by one, and I read every message. If this sounds like your setup, get in the queue — I'll write back.
— zpashafounder, xector
Get in the queue →launching soon
Request a sealed office.
We open soon, to a small group of teams first. Tell us where to write — we'll set up your machine and your first agent with you.