From request to operation
Every change goes through the same seven steps. AI speeds things up, but at two points a human always decides: the engineer who reviews the code, and you, before anything goes live.
- 01 01RequestYou
Described in the dashboard — as text, a voice note or a document.
- 02 02AI analysisAI
Checks the request against the existing code, fills in details and asks questions.
- 03 03DevelopmentEngineer + AI
Built in an isolated development environment, with test data only.
04ReviewEngineer An engineer reviews every change before it moves on. No approval, no release.
Human review 05Staging approvalYou You see the change in a preview and decide whether it goes live.
You decide - 06 06DeployPipeline
Built in the pipeline, rolled out as an immutable image, without downtime.
- 07 07OperationOmnisnap
Monitoring around the clock. Back to any earlier version in minutes.
The Schleuse: architecture of the one way in
Your environment has exactly one way to change it: the ticket through the Omnisnap DevOps Gateway. There is no direct write access to servers, database or code that bypasses the pipeline.
- 01Gateway
Every request becomes a ticket: logged, tied to a company and a person, never anonymous.
- 02Analysis against the source code
AI reads the request against your system's existing code and flags gaps, contradictions and risks before anything is built.
- 03Guardrails
Requests that would do harm, extract data or bypass protections are blocked by the gateway. An engineer decides borderline cases.
- 04Isolated development
Building happens in a separate environment with test data. Production data and secrets are out of reach from there.
- 05Review and approval
A named engineer reviews every change, the build scans code and dependencies, you approve on staging.
- 06Release
An immutable image is rolled out. Every version stays archived; a rollback takes minutes.
| Who | May | Never |
|---|---|---|
| Your users | File requests, answer questions, test and approve on staging | Change code or servers directly |
| AI | Analyse requests and write code in the isolated development environment | See production data or secrets, ship anything unreviewed |
| Engineers | Review, ship through the pipeline, intervene in operations with multi-factor sign-in | Deploy around the pipeline, access without a log |
| Nobody | — | A second way in: no shared admin accounts, no back door |
AI guardrails: how AI-written code is constrained and checked
AI writes a lot of our code, but never without a frame. The frame is enforced technically, not merely agreed.
The coding AI works with source code and test data only. Production data, customer data and secrets are out of its reach.
It knows our platform, its building blocks and rules, instead of writing code blind.
Type checks, build, dependency and security scans run on every change. What fails does not move on.
No AI code goes live before a named engineer has reviewed it and you have approved it on staging.
A dedicated environment per customer
No shared database, no shared storage. Every customer gets its own Google Cloud project in Zurich, with everything that belongs to it — layer by layer.
- Its own load balancer with managed TLS certificates
- Firewall: only what has to be reachable is reachable
- Denial-of-service protection right at the network edge
- Containers on Google Cloud Run
- Every instance in its own sandbox, separated by a virtualisation boundary
- Scales automatically with the load, in both directions
- MongoDB Atlas: three nodes in three zones, automatic failover
- Reachable only over private connections inside Google Cloud
- File storage and database encrypted (AES-256)
- Monitoring, logs and alerts around the clock
- Infrastructure as code: every environment described, versioned, reproducible
- Background work runs as separate, isolated jobs
- Google data centres in Zurich, in operation since 2019
- Titan security chip as root of trust, multi-layered physical access control
- Certified to ISO/IEC 27001, 27017, 27018, 27701, SOC 1–3, CSA STAR, BSI C5
Only reviewed code goes live
No release is built on a laptop. Whatever runs in production was built in the pipeline from versioned, reviewed code — and is bit for bit what you saw in the preview.
Codeversioned, every line traceable Reviewan engineer reviews and approves Buildin the pipeline, never locally Imageimmutable, archived Stagingpreview, your approval Rolloutstarts in parallel, takes over when ready Livemonitored around the clock
The new version starts next to the old one and only takes over once it is ready. Your users notice nothing.
A built image is never changed afterwards. What was reviewed is what runs.
Network, services and permissions are described and versioned, not clicked together by hand.
Updates run live; there are no maintenance windows. Larger migrations are announced weeks ahead.
Data model: flexible, not rigid
Your data is stored as JSON documents. Data models grow with your processes, without rigid migrations and without interruption.
A new field needs no migration of the whole table. Existing records stay valid, new ones carry the field.
Which fields are searched and sorted is defined per data type and versioned, like the code.
As the model does not have to be rebuilt, updates run live. Larger migrations are announced weeks ahead.
When you leave, you get the data in the same open format, without conversion, without loss.
Security in layers
No single mechanism has to hold everything. Each layer is secured on its own — if one gives way, the others stand.
TLS for every connection Firewall and load balancer per customer Database reachable only over private connections
Multi-factor sign-in for every access to our systems Only named engineers, only as far as operations and support require Every access to your environment is logged
Encrypted at rest (AES-256) and in transit (TLS) Stored in Zurich, separated per customer Development with test data only
Four-eyes principle: every change reviewed Security updates for libraries and platform continuously, no ticket needed Secrets never in source code; removed and renewed on every handover
Monitoring and alerts around the clock You are informed without delay, within 24 hours at the latest A written, blameless analysis after every critical incident
Beneath every layer sits the physical security of Google's Zurich data centres — a level hardly any in-house server room reaches.
AI with clear rules
AI plays two roles with us: it helps our engineers build, and it works inside your modules. For both: your data trains no models, and a human decides.
- Analyses tickets and checks them against the code
- Supports the engineer while building — on test data
- A human reviews every change before it goes live
- Google Vertex AI (Gemini) under enterprise terms
- Requests run through your own Google Cloud project
- Can be switched off per area if you want no AI there
No training on your data: not with us, not with Google AI in your processes is Gemini via Vertex AI, processed in Zurich, and only where your process provides for it The coding AI works with source code only, never with your customer data You know which modules use AI and for what
Sending a large LLM like Gemini the model instead of the data
When AI builds a report for you, it never gets to see your customer data. It gets the structure of your data and the question — and writes the calculation from that. The calculation then runs on your side, on the real numbers.
- Müller AG · East · CHF 4,820
- Keller GmbH · West · CHF 1,290
- Huber & Co · East · CHF 7,300
- Bernasconi SA · South · CHF 2,640
Backups, versions, recovery
A backup only counts once it can be restored. That is why we back up continuously, keep copies in tiers and test restoring regularly.
Source code in a complete version history, every shipped version as an immutable image.
Trash first; on top of that, storage retains deleted files for at least 7 days.
Restoring from backups is tested at least quarterly.
Single records, files or a complete state — usually within one business day.
Operations & SLA
Hosting, operations, monitoring and updates are included in the module price. What we guarantee is in your SLA — with figures we disclose to you regularly.
Software unusable or data at risk: priority over everything else, we work until operation is restored.
An important workflow is broken and there is no workaround.
Changes, improvements, minor bugs — delivered, not just acknowledged.
Availability, response and delivery times — at least quarterly, reviewed together once a year.
On request you see the database, storage and logs of your environment yourself.
Cause, impact and what we change — in writing, after every critical incident.
The highest level: in your own infrastructure
For companies whose production data must never enter our reach. A level few providers offer at all.
- A project of your own in the Google Cloud Zurich region
- Production access only by named engineers, as far as operations and support require
- Multi-factor sign-in, every access logged
- The Omnisnap Cloud runs in your infrastructure
- A production-to-development interface: we build and test with test data
- Our engineers never see your production data
- Same Schleuse, same features
Integrations in three levels
Your software does not stand alone. How deeply it connects to your existing systems is your choice, system by system.
Data is taken over from your system, regularly or at the push of a button.
Changes flow both ways; both systems stay current.
Datasets are compared continuously, differences explained and repaired.
Your code. No lock-in.
Your application belongs to you. The Omnisnap technology underneath comes with a perpetual, free licence — and everything runs in any Docker environment.
On request continuously into your GitHub or GitLab — every version that goes live.
Buildable codebase with infrastructure description, data as JSON, documents in the original — within 10 business days at most.
Up to four handover sessions with you or your new provider, 12 months of e-mail support.
For a code review, audit or due diligence, usually within 2 business days — secrets removed.
More than code
AI has made generating code easy. Running software has not become any easier. The difference lies in everything that comes after the first prompt.
OmnisnapVibe coding is tempting: fast, cheap, seemingly easy. But without a system underneath that is scalable, flexible and above all secure, the quick fix turns into an expensive risk — open vulnerabilities, lost data, outages and software nobody can maintain. We build just as fast, on a foundation that holds.
We take the time for your architects and your IT security: questionnaire, architecture session, audit copy or read access to an environment.
This page describes principles and guarantees. The SLA in your contract is binding; we disclose specific configurations as part of a review.