$ claude --resume> refactoring auth module, tests running...[lid closes][phone attaches · attach #4]> ...auth module: tests passing, writing migration 01 · laptop · running 41> refactoring auth module ...42● agent running · session live43$ region: EU 02 · lid closed device offline● session live in EU 03 · phone · attached 41> refactoring auth module ...42● agent running · session live43$ region: EU
Close the laptop. The session keeps running.
A persistent dev environment, hosted in the EU by a French provider.
Swap agents without changing your host.
- Signable DPA
- Connect your Claude or Mistral account
- Annual SEPA billing
- A human answers
What you just saw
The session runs on a dedicated Linux box.
The session survives disconnect by design: it detaches rather than dying. You reattach from iPad, phone, or PC and pick up at the same line, scrollback intact.
How it works →Disconnect is not loss. The session line never breaks; only device lines come and go.
Your agent, your account
Swap agents without moving. The environment stays.
Claude Code, Codex, or Mistral. The environment stays, the key changes. Your API spend stays on your own account, not ours. Switch vendors and your session never notices.
Bring what you already use →$ permia envagent: claude code key: yoursagent: codex key: yoursagent: mistral key: yours
Where that session ran
That session ran in the EU, on a French provider.
Permia operates under French law, no US entity in the chain. The CLOUD Act reaches US-owned providers even with an EU datacenter. Your agent calls its vendor directly. The diagram shows that traffic leaving the boundary, because it does.
How the CLOUD Act applies to dev clouds →your devices
laptop · phone
your persistent terminal
code · env · history stay here
your vendor account · direct
session live
the environment stays here
model vendor
the one you chose
what the agent sends travels to the vendor you chose, through your account, never through Permia
What your DPO signs
Documents before contact.
You should be able to read what you would sign before you talk to anyone. So here it is.
DPA
DPA, enforceable downstream to your own clients.
When their client asks where the code runs, your answer is a contract, in writing.
The 5-euro VPS question
The cheap VPS works. We ran one for years.
Tailscale, mosh, an agent CLI, a 5-euro box: for a single user on low-stakes work, that stack wins, full stop. Well-run single boxes run for years; yours probably will too. What it does not give you: anyone watching when the agent session crashes at 3am, backups someone has actually restored from, hardening that gets maintained after the weekend you set it up, a DPA your VPS provider will never sign for your client, and someone accountable at 11pm before a delivery. Every one of those hours is unbillable.
the math, in unbillable hours
- setup + hardening
- your weekend, then again after the reinstall
- upkeep, backups
- your evenings, monthly
- the 3am crash
- your morning after
Pricing, plainly
Dev line
For the daily work: persistence, BYOK, zero admin.
Solo from 348 EUR/yr · Pro · Team
See Dev pricing →Regulated line
Compliance you can sign: DPA, single-tenant infrastructure.
4,990 EUR/yr · Enterprise on quote
See the regulated tier →One annual invoice by SEPA. Every self-serve price is public, and everybody gets the same price for the same thing.
Questions we actually get
Why not a cheap VPS?
It works, and we said so two screens up. The honest difference is who carries the upkeep and who signs the paper. A VPS hands you the box and nothing else: the hardening, the verified backups, the watch over long-running agent sessions, and the accountability are yours, in unbillable hours. And when your client's lawyer asks for a DPA, the VPS has no answer. If your time is genuinely free and no client ever asks, keep the VPS.
Why not a free US dev cloud?
For plenty of work, a free sandbox is fine. Two things it does not give you. First, persistence on your terms: free tiers recycle idle environments, and a session that evaporates is the exact problem you came here to solve. Second, an answer for the client who asks where their code runs: for buyers in health, legal, or the public sector, a US-owned provider is an answer their compliance team cannot accept, whatever the price. Wherever your own company sits, what you buy here is residency you can put in writing.
What if Anthropic cuts my access?
Then Anthropic has cut your access, and your environment has not noticed. We never sell or proxy model access, so there is no Permia account for them to act on. Your session, your files, and your shell keep running; you put a Codex or Mistral key in place of the old one and continue.
Is the infrastructure multi-tenant?
On the Dev line, yes, with per-client isolation: dedicated provisioning per account, no shared sandboxes, documented hardening and anti-abuse. The regulated line runs single-tenant on dedicated footing, with no pooling at all. Which one you need depends on what your clients require.
How does session persistence actually work?
The session persists on the server side via a multiplexer, and the product page explains the mechanism because you should verify it. What you are paying for is everything around that persistence: a box that stays up and patched without your attention, reattachment that works from a phone on hotel Wi-Fi, backups someone has tested, a DPA, and a human who answers. Two limits: mosh runs over UDP, which some corporate firewalls block, so we keep a plain SSH path as fallback; and a phone screen is good for checking on an agent and nudging it along, not for a full day of work.
You've seen the walkthrough. See it live.
Close the laptop. The session is still there tomorrow.
A person in Paris answers, not a ticketing bot.
Full pricing is public. Download the DPA.