EU AI Act ready — Aug 2026 compliant

Your AI that learns you,
not from you.

Kai is a self-hosted AI assistant that adapts to how you think, work, and communicate — while keeping every byte of your data on your own servers.

By invitation only. We review every request personally.
Request received. We'll be in touch within 48 hours.
GDPR by architecture AES-256 encrypted Your data trains nothing By invitation only
Adaptive persona
Kai remembers. Kai adapts.
Kai is yours.
Every conversation teaches Kai more about how you work. Your tone, your expertise, your clients, your preferences — all stored locally in an encrypted vault that never leaves your server.
K
Kai's memory vault
5 learned traits · encrypted
Tone preference
Direct and concise. Skip pleasantries. Prefers bullet-ready answers over long explanations.
Expertise level
Advanced developer — TypeScript, React, Next.js, Node.js, PostgreSQL. No need to explain basic concepts.
Personal context
Solo founder running three ventures from Barcelona. Multilingual: EN, ES, DE, CA.
Domain knowledge
SEO consultant pivoting to AI/GEO services. Understands Ahrefs, GSC, structured data.
Communication style
Prefers copy-paste-ready commands over theoretical discussion. Wants execution, not explanation.
◆

Learns from every conversation

Kai picks up your vocabulary, your preferences, your recurring topics. After a week it feels like talking to a colleague who knows your business.

◇

Per-conversation personas

Switch Kai into "Spanish legal advisor" for contract reviews or "medical intake assistant" for patient processing. Each persona carries its own system prompt and expertise.

◈

Memory you control

View, edit, or delete anything Kai remembers. Export it as JSON. Wipe it with one click. Unlike cloud AI memory systems, this vault lives on YOUR hardware.

◉

Client-specific knowledge

For professionals: Kai remembers your clients' names, case details, treatment protocols, project specs. Context that makes every interaction faster and sharper.

Privacy router
Sensitive data never leaves.
Automatically.
Kai scans every message for sensitive content — health records, legal terms, financial data, personal identifiers — and routes it to local-only inference. No toggle to forget. No mode to switch. It just works.
Sensitive query detected
"Review my patient's treatment notes from last Thursday and suggest a follow-up plan"
"Check the NDA clause about non-compete for client García"
● Ollama (local) — zero data transfer
⇋
Same app
Same UI
General query
"Explain the difference between flexbox and grid layout"
"Draft a blog post about AI trends in 2026"
● Groq / Mistral EU — frontier quality
Everything it does
Built lean. Ships complete.
Under 50MB. Deploys with one command. No GPU required for the platform — just plug in your model provider and go.
Streaming chat
Real-time token streaming via WebSocket. Responses appear word by word, just like ChatGPT. Markdown, code blocks, syntax highlighting built in.
Live
Multi-model
Groq (fastest), Mistral (EU-native), Ollama (local). Automatic fallback chain. Force any provider per message. Swap models in .env, not in code.
Live
Installable PWA
Add to home screen on iPhone or Android. Offline-capable service worker. Feels native. No app store needed, no review process, instant updates.
Live
GDPR toolkit
One-click data export (Article 20). Instant purge API (Article 17). Audit trails on every message: who, when, which model, which provider. Consent banner included.
Live
Encrypted vault
Personal context stored AES-encrypted at rest. Your preferences, your clients, your domain knowledge — accessible only with your vault key.
Live
Docker deploy
docker compose up. That's it. Works on a Raspberry Pi, a €7 Hetzner VPS, or enterprise infrastructure. Same image, different .env.
Live
Document RAG
Upload PDFs, contracts, medical guidelines. Kai answers grounded in your documents. SQLite vector search — no external DB needed.
Coming soon
Voice input & output
Speak to Kai via browser microphone. Transcribed via Whisper. Responses read aloud via browser TTS — 100% local, no audio data sent anywhere.
Live
Encrypted vault backup
AES-256-CBC encrypted backups with scrypt key derivation. Download a .enc file, store it anywhere. Only your password can decrypt it. Restore in one click.
Live
Push notifications
Web Push via VAPID for reminders, meeting alerts, and task deadlines. Processed locally — no third-party push service has access to your notification content.
Live
Self-update system
Check for updates from within Kai. One-click update pulls latest code, rebuilds, and restarts — or ask Kai in chat to update itself.
Live
Hardware-grade lockout
Escalating brute-force delays modelled after secure elements. Exponential backoff from 30s to 24-hour delays. Makes PIN guessing mathematically infeasible.
Coming soon
AES-256 database encryption
SQLCipher encrypted database at rest. Key derived from your PIN via Argon2. Even physical access to the storage can't read your data.
Coming soon
Duress PIN & remote wipe
Emergency PIN shows fake empty state while silently exporting encrypted backup and wiping data. Remote wipe via authenticated API for lost devices.
Coming soon
Multi-user SSO
Team deployments with SAML/OAuth, role-based access, per-user audit trails. Each user gets their own persona and vault.
Coming soon
Questions & answers
Everything you need to know.
Can't find an answer? Email us and we'll get back to you.

Kai is a privacy-first AI assistant that runs entirely on your own hardware. It combines the speed of cloud AI with the security of local processing — sensitive conversations never leave your network. Think of it as your personal AI that you actually own and control.

Professionals handling confidential data — lawyers, doctors, therapists, accountants, consultants. Privacy-conscious individuals and small teams. Anyone who believes their conversations should stay theirs, not train someone else's model.

Three things. First, your data lives on your device — not on someone else's server. Second, sensitive queries (health, finances, legal, passwords) are automatically kept local and never sent to any cloud. Third, Kai is a full productivity suite — not just a chatbot — with tasks, reminders, projects, notes, meetings, brainstorming, email, and calendar all in one place.

Anything that runs Docker or Node.js. A Raspberry Pi 5, a €7/month Hetzner VPS, your office NAS, or enterprise infrastructure. No GPU required for the platform — local inference uses Ollama which works fine on CPU for smaller models.

Chat with AI, manage tasks and to-do lists, set recurring reminders, organize projects with custom AI personas, brainstorm ideas with AI-generated suggestions, take notes, manage meetings with AI briefs and recaps, read and reply to emails, view your Google Calendar — all from one interface.

Kai analyses every message for sensitive content (medical symptoms, financial details, legal questions, passwords, personal identifiers). Sensitive topics are processed locally by Ollama on your device. Everything else goes to Groq or Mistral for fast cloud inference. You always see which provider handled your message.

Kai learns your tone, vocabulary, expertise, and preferences from every conversation. This memory is stored in an encrypted vault on your server. You can view, edit, or wipe it at any time. Projects can have their own persona with custom system prompts, so Kai adapts its behaviour per context.

Yes. Dictation transcribes your speech to text using Whisper. Voice mode reads Kai's responses aloud using your browser's built-in speech synthesis — 100% local, no audio data sent anywhere.

Yes. Attach images, PDFs, code files, and text documents up to 10MB. The AI uses them as context for the conversation. Full document RAG (upload once, query forever) is coming soon.

Create a password-protected AES-256 encrypted backup of all your data. Download the .enc file and store it anywhere — USB drive, cloud storage, another server. Only you can decrypt it. You can restore from a vault backup at any time from Settings.

On your server. Period. Kai uses SQLite — a single file on your disk. No third-party cloud databases, no external analytics, no telemetry. Your conversations, tasks, notes, and files stay on hardware you physically control.

Yes — by architecture, not by policy. Data residency is wherever you deploy (your server in the EU). You have full Article 15 access, Article 17 right to erasure (one-click delete), Article 20 portability (JSON export). Consent flow built in. No data is shared with or sold to any third party.

Only the current message and recent context is sent to the cloud provider (Groq or Mistral). They do not store or train on your data. Sensitive queries are automatically blocked from cloud providers and processed locally. You always see which provider handled each message.

HTTPS encryption, CORS origin restriction, rate limiting, Helmet security headers, JWT authentication, escalating brute-force delays (exponential backoff modelled after secure hardware elements — up to 24-hour lockout after 140+ attempts), AES-256 encrypted vault backups with scrypt key derivation, and stripped technical fingerprints on all responses. No admin backdoors, no analytics, no tracking pixels.

Kai uses an escalating delay system modelled after secure hardware elements. 1–4 wrong attempts: no delay. 5 attempts: 30-second wait. 10–29 attempts: 30-second waits. 30+ attempts: delays double exponentially every 10 tries. After 140 failed attempts, each retry requires a 24-hour wait. This makes brute-force attacks mathematically infeasible — it would take decades to guess even a short PIN.

Yes. Kai supports AES-256 database encryption at rest via SQLCipher, with keys derived from your PIN using Argon2. Even if someone physically extracts the database file, it’s unreadable without your credentials. On top of that, all vault backups use AES-256-CBC with scrypt key derivation.

A secondary PIN that presents a clean empty state as if Kai has no data. Behind the scenes, it silently exports an encrypted backup to a secure location and wipes the local database. Designed for high-risk scenarios where you may be forced to unlock your device under pressure.

Yes. Remote wipe triggers full data destruction via an authenticated API call — useful if your device is lost or compromised. Factory reset drops all tables and regenerates a clean database. Both are available on managed and enterprise plans.

Yes. Full Local Mode routes every query to the on-device AI engine. Zero cloud traffic, zero outbound connections to AI providers. The device becomes fully air-gap capable. Enable it in Settings for maximum isolation.

Kai is designed for compliance with the EU AI Act (effective August 2026). As a self-hosted tool where the user controls all data and model selection, it falls under the lowest risk category. Audit trails, model transparency, and data governance are built in from the start.

Yes. In Settings you can export all data as JSON, create an encrypted vault backup, or delete everything with PIN confirmation. You can also delete individual conversations, messages, tasks, notes, or any single item. Ghost mode creates conversations that leave zero trace.

Clone the repo, copy .env.example, add your Groq API key, and run docker compose up. That's it. Kai runs on any Linux server, macOS, or Windows with Docker. Under 50MB, no GPU required.

No. The Kai platform runs on CPU only. For cloud inference (Groq, Mistral), no local compute is needed. For local inference with Ollama, smaller models (3B–7B) work fine on CPU. A GPU only helps if you want to run large local models (70B+).

Yes — that's how we run it ourselves. Raspberry Pi 5 with 8GB RAM and an NVMe SSD. The entire setup costs less than one month of most AI subscriptions. Accessible from anywhere via a Cloudflare Tunnel.

Create a Google Cloud project, enable Gmail and Calendar APIs, add OAuth2 credentials to your .env file. Then connect from Settings inside Kai. Kai uses OAuth2 — it never sees your Google password. You can disconnect at any time.

Kai is currently available by invitation. Submit your email and we'll review your request personally. We prioritise professionals in privacy-sensitive fields — healthcare, legal, finance, consulting — and teams with specific compliance requirements.

Kai will offer a free self-hosted tier with full core features and no telemetry. Premium tiers with managed hosting, advanced security (database encryption, duress PIN, remote wipe), and enterprise features will be available at launch. Pricing will be announced soon.

Yes. We'll deploy and maintain a private Kai instance on EU servers for you — SSL, custom domain, auto-updates, daily backups, and email support included. Your data stays on your dedicated instance. Request access to discuss your setup.

Your AI should know you.
Not expose you.

Kai is currently available by invitation. Request access and we'll set you up with a private instance.

Limited spots. We review every request personally.