We run it. Agents watch it. You build.
The developer experience, rebuilt around one idea: you review decisions; the platform does the work and writes down what it did. That's what easy to use means here.
Your repo in, your services running
Connect a git repository or hand us an OCI image. You describe intent; the platform proposes sizing, dependencies and placement, and ships it on your approval, inside one EU perimeter.
Your app, actually run
Failures troubleshot, capacity watched ahead of demand, scaling and placement handled across providers. Hosted is not the same as handled.
You set how much it does alone
Every agent action has a dial: acts alone, asks first, or never. Defaults start read-only; autonomy is granted per action, and every grant is on the record. This surface is next on the build; the list below is honest about it.
Everything it did, written down
Every action becomes a record as it happens: deploys, changes, recoveries, restores. One trail, many uses: how your app behaved, what compliance asks, what security wants to see. Immutable, timestamped, yours.
| Service | Placement | Replicas | Status |
|---|---|---|---|
| apighcr.io/hire-loop/api:4f1e3 | EUDE-FSN1Hetzner | 3/3 | Live |
| webghcr.io/hire-loop/web:2c90a | EUDE-FSN1Hetzner | 2/2 | Live |
| db-eu-central-01managed postgres 16 | EUFR-PAR1Scaleway | 1 + standby | Live |
| rank-modelmistral-small · open weights | EUFR-PAR2EU AI endpoint | on demand | Next |
The parts, with the risk we see in each.
The stack, plainly: Kubernetes on Hetzner compute, managed Postgres on Scaleway, OCI images in, rolling deploys out, one EU perimeter. unhale runs on itself. Here's what each part is, and the risk we write down instead of talking past:
Multi-service deploys to managed EU clusters; our own product is the first workload.Risk: this layer alone is a commodity; it's the entry, never the edge.
What it costs, why it broke, what we did: operational honesty as the product surface.Risk: being transparent about providers we don't own means owning their failures in front of you.
The core bet, not a bolt-on: agents that see logs, events and deploy history, explain failure in your words, and draft the fix. Read-only first; autonomy is earned action by action.Risk: others sell this to experts; our bet is removing the expert, which is the harder promise.
Karpenter-class scaling on European iron: the piece the hyperscalers have and Europe doesn't.Risk: EU provider APIs weren't built for it; this is a research project, not a checkbox.
The platform's actions, placements and recoveries, written down as they happen: residency history, subprocessor register, backup attestations.Risk: most of compliance is paperwork on your side; this covers the infrastructure-shaped part, and we say so instead of selling certainty.
Model calls routed to endpoints under European jurisdiction; open-weight models on EU iron where they're enough.Risk: the frontier models are mostly American, so we name the trade-off per workload instead of hiding it.
If you'd rather read code than adjectives: write to us and we'll talk shop. The risk register is the sales pitch.
Our side of the line, and yours.
The boundary, made explicit: what the platform and our engineers carry, and what stays with you. If a row surprises you, that's what the contact form is for.
What you pay, what we stand behind, and the questions you should ask.
Running is usage-based, with no seat licenses, and the billing units are ours, not the providers', so the bill stays portable. The record comes with the platform; we don't sell your own history back to you. You can predict the bill before you deploy.
Mostly in paperwork on your side: policies, risk assessments, auditor management. Roughly a fifth of a typical framework is infrastructure-shaped: residency, access, encryption, backups, logging. We run that part, and the record answers for it. Anyone who says a platform makes you compliant is selling something.
An EU region on a US provider keeps your data in Europe physically. The operator still answers to US law wherever the server sits (jurisdiction follows the company, not the datacenter), and your customer's lawyer knows the difference. Our stack is European companies operating European infrastructure, so the answer holds.
The platform brings you decisions to approve, not dashboards to read. Agents act within bounds you've approved and are read-only everywhere else: acting while you're away is the last capability to earn, not the first. We publish what's autonomous and what isn't.
Your services are containers on standard building blocks (Kubernetes, managed Postgres, object storage) on providers you could contract directly. Our billing units are ours, not theirs, so the bill stays portable and leaving is a migration, not a rewrite. We'd rather earn the staying.
We're early and carry no badge wall. unhale runs on itself, on Hetzner and Scaleway, and our own ISO 27001 path is on the roadmap; we publish where we are on it rather than implying we're further along.
When did production last wake you?
And how much of last month went to infrastructure instead of the product?
We haven't launched. This is the design-partner phase: what you tell us decides what gets built, and in which order.
A concrete offer: describe your stack: the services, the providers, what runs where, what keeps breaking. We'll map how we'd run it hands free: what the platform carries, what stays yours, and what it would cost. The map comes back to you, no pitch attached.