AI on your terms

Enterprise AI —
operated, not just installed.

Q: What does an AI readiness workshop include?

In two weeks we clarify where and how AI is used today, which data is affected, which use cases bring value and what your regulation requires. You receive a situation picture, a target architecture, prioritised use cases and a cost frame — whether or not you continue with Klyo afterwards.

Q: Can you build and run private AI on our own infrastructure?

Yes — that is our core. We size GPU capacity against real demand, set up Linux, Kubernetes, network segmentation and private inference — on-premise, in Swiss data centres or in the cloud — and deliver everything repeatably as infrastructure code. Capacity follows measured usage, not assumptions.

Q: Can you connect AI to our documents and systems?

Yes. Our engineering connects Klyo to your documents, intranet, repositories and business systems — with permissions checked and sources cited. We measure answer quality per use case before anything goes live, then automate recurring workflows or build governed agents once control is proven.

Q: Do you also train our employees?

Yes. Training for leadership and users is part of our advisory: how to use AI productively, which rules apply and why. The goal is adoption — a controlled path people prefer over the workaround.

Q: What does 24/7 operations include?

Depending on the tier: monitoring, updates and backups; high availability with 24/7 alerting and on-call; an SLA with defined response and recovery times, reported monthly; model changes without interruption and evidence for audits. One point of contact for platform, models, integration and evidence.

Q: We already have our own AI solution. Can you still help?

Yes — that is our “Secure existing AI” entry point. In about four weeks we place the control layer over what you have built — Copilot, own models or pilots — so that data protection, evidence and cost control apply immediately. On request we take over operations.

Q: How is an engagement priced?

By scope. Advisory and build are priced as projects, the Klyo platform as a licence and operations as a monthly service with SLA. After the first conversation you receive a concrete proposal — no black boxes.

Q: We already have Microsoft Copilot. Why do we need Klyo?

Copilot is strong inside the Microsoft world. Klyo covers what it structurally cannot: other providers, private and local models, uniform rules across all AI tools — and the operation of that environment. We do not replace Copilot; we place the control and evidence layer over it.

Q: Where is our data processed?

Where you decide: on your premises, fully isolated (air-gap), in ISO 27001-certified Swiss data centres, or in Azure, AWS and Google Cloud in Swiss and EU regions — chosen per data class and combinable. Public AI is only used where you allow it.

Your AI & infrastructure partner from Switzerland — with our own platform at the core.

Your people already use AI. Klyo makes sure they do it with your data, under your rules and under your control — and that someone is accountable for operations once the pilot is over.

Book a conversation How we work

Your people

Klyo Control • Evidence • Operations

Public & private AI

Secure Sensitive data is detected before it leaves the company.
Sovereign On-premise, in Swiss data centres or in Azure, AWS and Google Cloud in Swiss and EU regions. You decide where your data is processed.
Auditable Every request logged — ready for audit, data protection and cost centres.
Operated Advisory, build, engineering and 24/7 operations from one team in Switzerland.

Your situation

AI is already in use in your company. The only question is whether you are steering it.

Nine out of ten Swiss employees use AI at work. Almost one in three does so through a private account — with customer data, contracts and internal knowledge. That is not misconduct. It is the consequence of missing alternatives.

29% use AI at work through private accounts — no corporate access, no log, no deletion concept. EY Switzerland AI Survey 2026

33% report higher AI costs than expected — because every department buys its own tools. Bitkom Research 2026

51% call data processing in Switzerland or the EU business-critical. EY Switzerland AI Survey 2026

What you may recognise What lies behind it

An AI usage policy exists — on paper.: Without technical enforcement it is a request. Internal audit will see it the same way at the next review.
Copilot is licensed, but not every case is covered.: Confidential data, domain knowledge and automation need other paths — and one place that treats all paths the same.
An AI pilot went well — and then faded.: The technology was there, an operating model was not. Nobody was responsible when the first model needed updating.
Your data protection officer asks about data flows to language models.: The revised Data Protection Act requires transparency and control over every recipient — an AI service is one.

Threat Vector Analysis

Data Exfiltration Vectors

Direct PII leakage via unfiltered prompt context windows
Indirect extraction through multi-turn conversational probing
Shadow AI: corporate data pasted into private accounts
Cross-tenant data leakage in shared inference endpoints

Prompt Injection Taxonomy

Direct injection: overriding system prompts via user input
Indirect injection: payload delivery through retrieval context (RAG poisoning)
Multi-modal injection: steganographic payloads in image/audio inputs
Tool abuse once agents are allowed to act in systems

What Auditors Ask (revised FADP)

Processing register: which AI services receive personal data?
Transparency towards data subjects
Processors and sub-processors, including model providers
Cross-border disclosure and its safeguards
Deletion concept for prompts and responses

What Klyo does for you

One access point for all AI. Your rules. Your data stays where you decide.

For your employees, Klyo is a modern AI workspace — with the best models, fast and without detours. For your IT and compliance teams, it is the one place where everything converges, is controlled and logged.

Your people Chat · Apps · Assistants

Klyo

Who may do whatYour login, your roles
What is protectedPeople, contracts, secrets
What happens to itMask, reroute, block
Which modelBy sensitivity, quality, cost
No way around itEnforced in the system
What is evidencedAudit, data protection, cost centres

On your premises, air-gapped or in a Swiss data centre

Public AIOpenAI, Anthropic, Google, Microsoft
Your private AIOwn models, your data centre
Your company knowledgeDocuments, M365, internal systems

Secure

Personal data, secrets and confidential content are detected and handled by your rules — before they reach a model.

Fast

The best models in one interface. No detours, no waiting for approvals. Adoption instead of workarounds.

Sovereign

Confidential content stays in your infrastructure or in Switzerland. Public AI only where you allow it.

Auditable

Who, when, which data, which model, which rule. Exportable for audits, chargeable by cost centre.

Where Klyo runs — your choice per data class

The same policies and the same evidence everywhere. You decide per data class — and can combine all options.

Highly sensitive

On-premise & air-gapped

On your hardware, fully isolated if required — operated by us. Not a byte leaves the network.

Confidential

Swiss data centres

Klyo capacity in ISO 27001-certified data centres in Switzerland or with Swiss sovereignty providers.

Standard workloads

Public cloud

Microsoft Azure, AWS and Google Cloud in Swiss and EU regions — on managed Kubernetes, with the same rules as everywhere else.

Hybrid

All at once

Klyo routes each request by data class. Sensitive content stays private, everything else may use the cloud.

Runs in Swiss cloud regions

Microsoft AzureSwitzerland North · Switzerland West
Amazon Web ServicesEurope (Zurich)
Google Cloudeurope-west6 (Zurich)

Models from Azure OpenAI and Amazon Bedrock can be connected as providers — under the same policies.

Architecture Deep-Dive & Tech Stack

Stack

Frontend: Next.js 15 + React 19 (SSR)
API: FastAPI (Python 3.12), async, auto-docs
DB: PostgreSQL 16 + TimescaleDB
Cache: Redis 7 (sub-ms, task queues, rate limits)
LLM: LiteLLM (100+ providers unified API)
Monitoring: Prometheus + Grafana
Auth: JWT + bcrypt (cost 12) + TOTP

Security Pipeline

NER-based PII detection (spaCy + custom transformers)
Prompt injection classifier (fine-tuned DeBERTa)
Regex + semantic topic boundary enforcement
Configurable redaction strategies (mask / hash / tokenize)
SQL injection prevention via SQLAlchemy ORM
XSS prevention via React built-in escaping
Zero third-party analytics or tracking

Routing & Infra

4 strategies: cost, quality, latency, balanced
Circuit-breaker (>50% fail in 5min → auto-demote)
EMA-smoothed per-model performance tracking
SSE streaming with segment-based Markdown rendering
Kubernetes-native: horizontal scaling, zero-downtime
Sub-50ms security overhead per request
Air-gapped on-premise deployment ready

Four services, one accountability

We do not just install. We advise, build, engineer and operate.

AI projects rarely fail on technology. They fail at the seams between software, infrastructure, integration and operations. At Klyo those seams do not exist — one accountability, one point of contact, in Switzerland.

Advisory

Understand before building

AI readiness assessment in two weeks
Governance and risk assessment
Target architecture and total cost
Provider and sovereignty decision
Training for leadership and users
Preparation for audit and data protection

Build

Production-ready in weeks

Klyo installation and hardening
Integration with your login and roles
Private models on-premise, in Swiss data centres or in Azure, AWS and Google Cloud
Network, segmentation, air-gap
GPU sizing against real demand
Repeatable as infrastructure code

Engineering

AI that knows your business

Connection to your documents and systems
Domain assistants built on Klyo, connected to your systems
Quality measurement instead of gut feeling
Process automation
Governed agents with approvals
Integration into existing tools

Operations 24/7

So that it runs — permanently

Monitoring and alerting
Round-the-clock on-call with SLA
Patching and security updates
Model changes without interruption
Cost and capacity management
Evidence for every audit

Use cases

What your teams do with it on day one — and what is possible after six months.

We start with what brings immediate value and carries little risk. Automation and agents follow once control and quality are proven.

Use case	What happens	For whom	When
Secure AI workspace	All employees use the best models through one access point — with data protection that applies automatically.	All departments	Day 1
Process confidential documents	Contracts, HR data, financial reports — processed in your private AI, without a byte leaving the building.	Legal, HR, finance	Week 2
Make knowledge usable	Questions to your documentation, intranet and repositories — with sources cited and permissions checked.	Business units, support	Month 1–2
Cost control	Quotas per team, routing to the cheapest suitable model, chargeback by cost centre.	IT management, finance	Month 1
Private AI infrastructure	Open models on your own GPU servers or Kubernetes cluster — sized against real demand, monitored and operated by us.	IT infrastructure, platform teams	Month 2–3
Process automation	Recurring workflows — ticket triage, reports, data reconciliation — run under control and are logged.	Operations, process owners	Month 4–6
Governed agents	AI completes tasks in systems — only within defined roles, with approval and evidence.	Operations, business units	From month 6

Timings typical; they depend on environment and data situation.

By industry

Where we see the most value in regulated and knowledge-intensive organisations.

Public sector

Federal, cantonal, municipal

Answer citizen enquiries faster — drafts with sources, approved by staff
Summarise consultation responses, submissions and case files
Search laws, ordinances and internal directives with citations
Prepare tenders and evaluate offers against requirements

Air-gap capable, official secrecy and archiving rules enforced, Swiss operator.

Finance & insurance

Banks, insurers, asset managers

Review contracts and policies against internal rules
Summarise regulatory changes and their impact
Support claims handling and client correspondence
Check onboarding and KYC documents

Outsourcing requirements and sub-processors documented and evidenced.

Healthcare

Hospitals, clinics, care

Draft discharge letters and reports from clinical notes
Summarise patient histories for handovers
Answer questions on internal guidelines and procedures
Prepare coding and billing documentation

Patient data never leaves the network — private models only.

Industry

Manufacturing, energy, engineering

Search manuals, specifications and service reports
Support service technicians with troubleshooting
Draft quality and incident reports
Onboard new staff with the knowledge of experienced colleagues

Design and process knowledge stays on your infrastructure.

IT & infrastructure

Linux, Kubernetes, GPU

Run private models on your own GPU servers or Kubernetes
Size GPU capacity against measured demand
Build AI platforms repeatably as infrastructure code
Segment networks and operate fully air-gapped

Linux, Kubernetes and networking down to kernel level — our home turf.

Legal, HR & back office

Every organisation

Compare contracts and flag deviations
Answer employee questions on policies and regulations
Draft letters, minutes and reports
Translate between German, French, Italian and English

Useful for every department from day one.

Operations

The pilot is the easy part. We stay when it is over.

Models change every few months. Security patches, capacity, cost and evidence change daily. Klyo operates your AI environment like any other business-critical platform — with availability targets, on-call and clear accountability.

24/7 On-call and alerting Swiss operations team, response times contractually defined by SLA.

SLA Availability, response, recovery Contractually defined, reported monthly.

1 Point of contact For platform, models, integration and evidence. No hand-offs between vendors.

Tier	Scope	For whom
Foundation	Monitoring, updates, backup, business-hours support.	First production use, non-critical use cases.
Business Critical	High availability, 24/7 alerting and on-call, SLA, capacity management, incident response.	AI is part of daily business.
Regulated	Evidence, audit exports, policy lifecycle, support during audits and regulatory reviews.	Banks, insurers, healthcare, public sector.
Model Lifecycle	Benchmarking of new models, changes without interruption, regression tests, continuous quality measurement.	Anyone who does not want to evaluate every quarter themselves.
Agent Operations	Tool inventory, agent identity, approval processes, runtime review.	When AI is allowed to act.

Tiers can be combined; terms per agreement.

How to start

Three entry points. Each in weeks, not quarters.

You do not have to decide everything at once. Start where your need is greatest — every entry point leads into the same controlled platform.

Entry 1

AI readiness workshop

2 weeks

Where is AI used today — and how?
Which data is affected?
Which use cases bring value?
What does your regulation require?

Outcome: situation picture, target architecture, prioritised cases, cost frame.

Entry 2

Secure AI pilot

6–8 weeks

Klyo in your environment, on your login
One or two production use cases
Real data, measured quality
Security review and evidence

Outcome: production platform, proven value, basis for decision.

Entry 3

Secure existing AI

4 weeks

You have already built — Copilot, own models, pilots
We place the control layer on top
Data protection, evidence, cost — immediately
We take over operations on request

Outcome: what you built becomes auditable and operated.

Weeks 1–2 Understand Workshop, data situation, use cases
Weeks 3–4 Design Architecture, policies, location
Weeks 5–8 Build Installation, integration, first case in production
Weeks 9–12 Measure Quality, usage, cost, security review
From week 12 Operate SLA, on-call, expansion

Timings typical for mid-sized environments.

Technical facts

Enterprise-grade from day one — so your IT and security teams can sleep.

What your IT department wants to know before it lets a control system into its network.

Identity and access: Integration with your existing SSO. Roles and permissions from your directory. No second user administration.
Data protection: Real-time detection of personal data, secrets and confidential patterns — before transmission to a model. Masking, rerouting or blocking per policy.
Models: Public providers, open models and your private models in one interface. Providers interchangeable at any time, rules stay. No lock-in.
Enforcement: Policies apply not only in the browser but at system level — including for applications and processes that try to send around the control.
Location: On your premises, fully isolated (air-gap), in ISO 27001-certified Swiss data centres, or in Azure, AWS and Google Cloud in Swiss and EU regions. Your choice per data class — also combined.
Evidence: Complete log per request: user, time, data class, model, policy, outcome. Exportable for audit, data protection authorities and cost allocation.
Operations: High availability, monitoring, defined recovery objectives, patching, model lifecycle. SLA-capable.

Detailed architecture and security documentation available under NDA.

The platform in detail

What is inside the Klyo platform — for IT and security teams.

Zero-Trust AI Firewall

Real-time PII detection & redaction, prompt injection prevention, file blocking, and custom rules — with a live firewall audit dashboard showing action breakdown, daily trends, and top triggered rules.

NER-based entity extraction with custom transformer models. DeBERTa-v3 prompt injection classifier running sub-50ms inference. Configurable redaction pipelines: mask, hash (SHA-256), or reversible tokenization with HSM-backed key rotation. Pattern-based rules via regex, keywords, and semantic matching. SQL injection, code injection, and prompt injection blocking. Firewall dashboard with action breakdown (block, redact, alert, log), daily trend analysis, searchable event log filterable by action type over configurable time periods (7/14/30/90 days).

Full Compliance Engine

Immutable audit trails with structured config diffs, conversation review queues, mandatory reason fields for critical actions, and export-ready logs for internal audit, data protection and cost centres.

Append-only Merkle tree audit log with cryptographic tamper detection. Structured JSON config diffs with red/green before-after panels. Mandatory reason fields on user deactivation and password reset stored in audit log. Soft-delete architecture: users delete chats from their view, admins retain full history. Message edit & regenerate with full edit history preserved. CEL-based policy evaluation engine. Automated DPIA generation with risk scoring. CSV/JSON export across all admin modules with metadata envelope (timestamp, exporting admin, applied filters, record count).

Intelligent Model Routing

100+ LLM providers via a unified interface. Four routing strategies — cost, quality, latency, and balanced — with circuit breaker auto-failover, per-model performance tracking, and one-click provider setup.

Unified inference API normalizing provider-specific schemas (OpenAI, Anthropic, Google, Groq, Mistral, DeepSeek, xAI, Cohere, Together AI, Perplexity, Azure OpenAI, AWS Bedrock, Ollama). Weighted scoring matrix (latency / cost / quality) per request class. Circuit breaker pattern: models with >50% failure in 5min auto-demoted, ranked last as fallback. EMA-smoothed per-model latency, tokens/sec, and success rate tracking. Quality-driven routing using user feedback scores (min 5 ratings threshold). "Auto (Smart Routing)" synthetic option in model picker with per-message model attribution. Routing decision log persisted with full scoring breakdown. Quick Add Provider: paste API key, all models auto-configured. SSE streaming pipeline with segment-based markdown rendering and live Prism.js syntax highlighting.

Enterprise Security

TOTP-based 2FA with org-wide enforcement, SSO via LDAP/AD and OIDC, API key authentication, encrypted provider credentials, Redis-backed rate limiting, and brute force protection.

TOTP-based 2FA with QR code provisioning and organization-wide enforcement toggle. SAML 2.0 + OIDC federation with auto-provisioning (Keycloak, Azure AD, Okta, Google). LDAP/Active Directory integration with mapped attributes. API key auth via klv_ prefixed keys — SHA-256 hashed, expiration, usage tracking, user/admin revocation. Provider credentials encrypted at rest via Fernet (AES-128-CBC + HMAC-SHA256) in PostgreSQL; master key from Docker secrets (/run/secrets/klyvex_key). Optional HashiCorp Vault Transit engine integration. Redis-backed per-user/per-group rate limits (requests/hr, requests/day, tokens/day). Progressive brute force protection with configurable escalation to full account lockout. Global auth versioning: policy changes invalidate all sessions. Automated email alerts for lockouts, password resets, and new accounts.

Cost Intelligence & Analytics

Per-request cost tracking, admin cost dashboard with department & model breakdowns, user-facing usage meters with color-coded progress bars, and inline feedback for continuous model quality evaluation.

Per-request cost capture: prompt tokens, completion tokens, USD cost stored alongside every message. Flexible pricing engine configured per-model in YAML (cost_per_1k_input, cost_per_1k_output) with built-in fallback rates for GPT-4o, Claude, and Gemini. Local Ollama models default to $0. Admin cost dashboard: summary cards, daily trend chart, breakdown by department/model/user, filterable by time period (7/14/30/90 days). Chargeback-ready per-department and per-user data. REST API: GET /admin/costs?days=N. Cost intelligence panel: per-model cost trends, routing savings estimates, monthly projections. Inline feedback popup with thumbs up/down, predefined tags, and free-text. Admin ratings dashboard with per-model approval rates and tag frequency analysis. User-facing usage bars: color transitions at 70% (amber) and 90% (red) with auto-refresh.

Role-Based Access (LLM Groups)

Define which AI models each team can access. Flexible multi-group assignment, default groups for fallback, per-group routing strategies, and rate limits — with a full admin UI for real-time management.

LLM Groups: group-based permissions defining which models each team accesses. Flexible assignment with multi-group membership — users inherit the union of all allowed models. Default groups as fallback for users not explicitly assigned. Per-group routing strategy configuration with custom balanced weights and fallback preferences. Per-group rate limits: rate_limit_per_hour, rate_limit_per_day, max_tokens_per_day. Users in multiple groups get the most permissive limit. Zero-config backwards compatibility: without groups, all users see all models. Admin UI with real-time member assignment and model selection.

Plugin Ecosystem

Extend Klyo with hook-based plugins. Auto-discovery from the plugins directory, a curated plugin store, and a simple Python API — build custom middleware for content filtering, rate limiting, and more.

Three plugin hooks: pre_request (before firewall, can block/modify), post_firewall (after firewall, before LLM), post_response (after LLM response, can modify). Auto-discovery: drop a Python file in gateway/plugins/ extending BasePlugin. Plugin store with curated options: content filter, rate limiter, audit logger, PII detector, cost tracker, response cache. PluginContext provides message, user_id, conversation_id, metadata. PluginResult supports blocked, block_reason, modified_message. Enable/disable via admin UI without restart.

White-Label & Branding

Deploy Klyo under your own brand — custom company name, logo, colors, login background, and domain. Feels like an internal tool, not a third-party service.

Streaming & Local Inference

Real-time token-by-token streaming via SSE with live syntax highlighting. Optional GPU-accelerated local inference via Ollama for maximum data sovereignty and zero API costs.

SSE pipeline: LLM responses streamed as JSON-encoded SSE chunks, preserving newlines and special characters. Segment-based Markdown rendering: incoming content split into text segments, complete code blocks (full Prism.js highlighting), and partial streaming blocks (live preview with progress indicator). Docker Compose GPU override for NVIDIA GPUs: allocates all GPUs to Ollama, increases parallel handling to 4, keeps models in VRAM for 30min. VRAM optimization with configurable overhead reservation. CPU-only mode works out of the box; GPU is opt-in.

For security teams

The difference between a rule and an enforcement

Most AI gateways control the traffic that voluntarily passes through them. Klyo additionally controls what a system is allowed to send out at all — independent of the application, directly in the Linux kernel.

Kernel-Level Egress Enforcement

eBPF programs hook deep into the Linux network stack and inspect every outbound LLM request before it leaves the host. Policy decisions execute below userspace — there is no application-layer code path an attacker can bypass.

BPF programs attached to socket and traffic-control hooks (cgroup_skb, tc-egress) inspect SNI and IP at packet level. Allow/deny decisions backed by BPF maps populated from policy state. Privilege isolation: enforcement plane runs with CAP_BPF and CAP_NET_ADMIN; gateway userspace remains unprivileged. Compatible with stock kernels ≥ 5.15. Compiled with libbpf CO-RE; portable across host distros without recompilation.

Compromise-Resistant by Design

Privilege separation isolates the enforcement plane from the application. Even with full RCE on the gateway, an attacker still cannot reach an LLM provider — the kernel-resident policy is the last line of defense, and it does not trust userspace.

Cryptographically Verifiable Audit

Every allow, deny, and policy mutation is written to an append-only, hash-chained audit log. Auditors can prove what reached an LLM provider, what was blocked, and when policy changed — without trusting the gateway process.

Observe Mode

Logs every LLM-bound flow without blocking. Drop-in deployment, zero risk to existing traffic. Ideal for the discovery phase — see exactly which models, endpoints, and providers your applications touch.

Enforce Mode

Blocks unsanctioned LLM egress at the packet level. Allow-listed providers and policy-approved routes only. Designed for self-hosted and bare-metal Kubernetes deployments where the kernel is yours to govern.

Our own product

Built by us. Improved every week.

Klyo is not resold software. We develop the platform ourselves — Gateway, Context Engine and Marketplace — with more than 100 releases so far. Every improvement reaches your installation through our operations.

100+ Releases shipped across the platform

44 Releases in the last 90 days

3 Products in active development

10+ Cloud LLM providers preconfigured

Sep 2026

Tamper-evident audit log

Every request is recorded in a tamper-evident log with immutable off-site storage — evidence your auditors can rely on.

Sep 2026

Enterprise single sign-on

Multiple identity providers with group-based access and roles — Entra ID, LDAP and more, hardened with PKCE and strict token validation.

Jul 2026

Klyo Prompt Coach

Suggests a stronger prompt before you send, so every query lands better on the first try — higher-quality answers with fewer wasted calls.

New product

Klyo Context Engine

An ML-based classifier that became our default PII engine — multilingual across EN/DE/FR, with calibrated confidence on every signal.

Jul 2026

One-Command Kubernetes

Deploy the full Klyo stack to Kubernetes with a single Helm command, plus one-step production HTTPS via a bundled reverse proxy.

May 2026

Automatic Model Fallback

When a provider errors, Klyo reroutes to a healthy model automatically — no dropped requests, no manual intervention.

Explore the full changelog

FAQ

Frequently Asked Questions

What customers ask us before the first conversation.

What exactly is Klyo — software or a service provider?

Both, deliberately. Klyo is an AI and infrastructure partner with its own product: we advise, build, integrate and operate enterprise AI — from GPU servers and Kubernetes to the models — and our own control platform sits at the core. For your employees it is a modern AI workspace; for IT and compliance it is the one place where all AI usage is controlled and logged.

What does an AI readiness workshop include?

Can you build and run private AI on our own infrastructure?

Can you connect AI to our documents and systems?

Do you also train our employees?

What does 24/7 operations include?

We already have our own AI solution. Can you still help?

How is an engagement priced?

We already have Microsoft Copilot. Why do we need Klyo?

Where is our data processed?

How does Klyo protect sensitive data?

Which regulations does Klyo help with?

For most Swiss companies the revised Data Protection Act is decisive: processing register, transparency, control over processors and cross-border disclosure. Klyo logs who sent which data to which model, when and under which policy — exportable for internal audit, data protection and supervisory authorities. Where the EU AI Act or sector rules (finance, healthcare, public sector) apply, the same evidence supports them.

How quickly can we start?

Who operates Klyo after the pilot?

Does Klyo work with any AI model?

What happens if a system tries to bypass the control?

Can Klyo run in Azure, AWS or Google Cloud?

Is Klyo suitable for public administrations?

Yes. Klyo can run fully air-gapped on your own infrastructure or in Swiss data centres, operated by a Swiss team. Every request is logged for archiving and oversight, and rules on official secrecy and data protection are enforced technically — not just on paper. Typical first use cases are answering citizen enquiries, summarising submissions and searching laws and internal directives.

The next step: one conversation.

Where does your AI usage stand today, what is your biggest risk, what is your biggest opportunity? Afterwards you will know which entry point is right for you — or whether you need none.

Book a conversation

Directly with our founders — pick a time that works for you via Cal.com.

or send us a message

Enterprise AI — operated, not just installed.

AI is already in use in your company. The only question is whether you are steering it.

One access point for all AI. Your rules. Your data stays where you decide.

Secure

Fast

Sovereign

Auditable

Where Klyo runs — your choice per data class

On-premise & air-gapped

Swiss data centres

Public cloud

All at once

We do not just install. We advise, build, engineer and operate.

Advisory

Build

Engineering

Operations 24/7

What your teams do with it on day one — and what is possible after six months.

By industry

Public sector

Finance & insurance

Healthcare

Industry

IT & infrastructure

Legal, HR & back office

The pilot is the easy part. We stay when it is over.

Three entry points. Each in weeks, not quarters.

AI readiness workshop

Secure AI pilot

Secure existing AI

Enterprise-grade from day one — so your IT and security teams can sleep.

The platform in detail

Zero-Trust AI Firewall

Full Compliance Engine

Intelligent Model Routing

Enterprise Security

Cost Intelligence & Analytics

Role-Based Access (LLM Groups)

Plugin Ecosystem

White-Label & Branding

Streaming & Local Inference

The difference between a rule and an enforcement

Kernel-Level Egress Enforcement

Compromise-Resistant by Design

Cryptographically Verifiable Audit

Observe Mode

Enforce Mode

Built by us. Improved every week.

Tamper-evident audit log

Enterprise single sign-on

Klyo Prompt Coach

Klyo Context Engine

One-Command Kubernetes

Automatic Model Fallback

Frequently Asked Questions

The next step: one conversation.

Message Sent!

Enterprise AI —
operated, not just installed.