Our models.Our gateway.Your AI.
In-house models and one OpenAI-compatible gateway — a sovereign AI platform from the Netherlands, and our products that run on it.
THE PLATFORM
We build the intelligence ourselves.
One platform underneath — in-house model families and an OpenAI-compatible gateway. A full suite of products on top.
Five in-house model families and one gateway, woven into a single network that every product draws on — no rented intelligence, no vendor lock.
THE FOUNDRY
Intelligence we forge ourselves.
Five in-house model families — language, speech, voice, vision, retrieval — trained and tuned in-house for production.
Our language model family — reasoning, generation and chat, trained and served in-house.
Our text-to-speech family — sub-second streaming synthesis and consented voice cloning.
Our speech-to-text family — fast, accurate transcription across languages.
Our vision family — OCR, background removal, image generation and visual understanding.
Our retrieval family — embeddings and reranking that make production RAG precise.
external providers
THE CONSTELLATION
Many products, one source of light.
Each product is a planet in its own color — all in orbit around the same sovereign AI platform.
Qevron
Qevron is the gateway and control plane for the whole platform. Point your code at one OpenAI-compatible endpoint and reach Arpanet's first-party models plus 43+ external providers, with intelligent routing, caching and a built-in tokenizer. Watch every token through live monitoring and cost analytics, all from one console.
- OpenAI-compatible API across 43+ providers + first-party models
- Intelligent routing, caching, websearch & think-split plugins
- Live monitoring, cost analytics, multi-tenant
Calleague
Calleague runs realtime STT → LLM → TTS over real SIP telephony, with production RAG and a visual workflow editor. Run inbound support and outbound campaigns through one multi-model gateway. Deploy on-prem or as SaaS, with the data sovereignty enterprises require.
- Realtime STT → LLM → TTS pipeline
- SIP telephony, DTMF, inbound & outbound campaigns
- Production RAG (hybrid BM25 + vector + rerank)
Blab
Blab is the studio surface for our Blab TTS family. Stream speech in under a second, clone voices with consent, and dub across languages while preserving the original soundtrack. 40+ languages, sixty-four voices, voice agents and batch jobs — wired into a REST API, a TypeScript SDK, MCP and signed webhooks. Voices so real, we require consent.
- Sub-second streaming TTS
- Consented voice cloning
- Soundtrack-preserving dubbing
Talky
Talky is the sovereign communications workspace: video conferencing in three room types, messaging, an ArpBoard whiteboard, calendar, notes, file sharing, screen share and recording. AES-256 encryption, lobby and room-lock, and RBAC keep every session under control. Deploy on-prem, stay KVKK-aligned, and reach your team on web, iOS, Android and desktop.
- Video conferencing (3 room types), messaging / DM
- ArpBoard whiteboard, calendar, notes, file sharing
- Screen share & recording
SetGet
SetGet is the AI-native work OS that replaces the stack you've duct-taped together. Projects, a wiki, realtime chat, 27+ dashboard widgets and AI agents share one workspace. View work as Board, List, Spreadsheet, Gantt or Calendar, migrate in from Jira, Linear, Asana, ClickUp or Monday, and self-host or run it fully isolated with SSO, SAML, MFA and audit logs.
- Projects + wiki + realtime chat + dashboards (27+ widgets) + AI agents
- Board / List / Spreadsheet / Gantt / Calendar views
- Migrate from Jira, Linear, Asana, ClickUp, Monday
Sosyabot
Sosyabot turns a single brief into designs, videos, posts and replies across 14+ networks. A Canva-like multi-page editor and autonomous agents do the heavy lifting, smart scheduling picks the moment, and a deterministic brand-lock keeps everything unmistakably yours. Approvals, analytics and billing close the loop from idea to attribution.
- One brief → designs, videos, posts, replies
- Canva-like multi-page editor & autonomous agents
- 14+ networks, smart scheduling, analytics
Pickerman
Pickerman collects across news, RSS, APIs, video, Telegram, social, galleries and sitemaps, then extracts IOCs and named entities and clusters the same event from many feeds. Its AI synthesis fuses three or more sources into a single story, while a real-time operator console and editorial workflow keep humans in command. Self-hosted, no Docker required.
- Multi-source collection: news, RSS, API, video, Telegram, social, gallery, sitemap
- IOC + entity (NER) extraction
- Same-event clustering (embeddings + kNN)
Solab
Solab turns meetings, calls, lectures and voice notes into structured, searchable notes. Record in the browser or bring audio, video, a media link or a PDF; Solab transcribes — timestamped, multilingual and speaker-aware — then builds a note you can read five ways: Overview, Analysis, Transcript, Knowledge map and Chat. Every claim links back to the exact moment it was said, and semantic search spans your whole workspace — all on our in-house Solab speech-to-text family, through the Qevron gateway.
- Capture anything: browser recording, audio & video upload, media-URL import, PDF & documents
- Timestamped, multilingual, speaker-aware transcripts
- Read a note five ways: Overview, Analysis, Transcript, Knowledge map, Chat
Harbi
Harbi rebuilds news around the event, not the article: dozens of stories about the same development merge into one permanent page that is versioned as the story evolves — nothing is silently rewritten. Every factual sentence is bound to the sources it rests on; outlets copying the same agency wire count as one root, so “17 outlets” never masquerades as “17 confirmations”. Automated quality gates check source coverage, number and date consistency and risky language before publication, corrections are published visibly, and audio bulletins are read by a clearly labelled synthetic voice. There is no infinite feed and no clickbait — the day has an end, and Harbi measures whether you are informed, not how long you scroll.
- Event pages, not articles: one permanent, versioned page per development
- Sentence-level evidence — every factual claim is bound to its sources
- Independent-source counting: agency copies collapse into one root
VidiVidi
VidiVidi replaces the personal feed with a single shared broadcast: four fixed daily shows plus DÖNÜM, the special session that convenes when something big breaks — all on one clock, so everyone hears the same sentence at the same moment. A persistent cast of synthetic hosts, each with a stable role, a consistent way of thinking and their own voice signature, debates the agenda around the table. Listening is free; members leave questions for the table to pick up on air. And it is honest by design: synthetic is labelled synthetic, prepared is never sold as live, every number on screen is a real measurement, and a retracted broadcast is retracted everywhere — audio, transcript and clips.
- One shared broadcast on one clock — join late and you join at the common point
- Persistent synthetic cast with stable roles and voice signatures, always labelled as synthetic
- Four fixed daily shows plus DÖNÜM, the breaking-news session
Kimene
Kimene holds your pools — accounts, models, hosts, projects, roles and tools — with no permanent binding between them: the match is made only while work runs, and it is recorded. Routing is hard eligibility first, weighted scoring second; the chosen candidate and every rejected one land in an immutable decision record, component by component, penalty by penalty. Work then executes on your real machines through the providers’ own CLIs, inside agent-managed tmux sessions — close the browser, redeploy the server, drop the SSH: the session stands, and an operator can take the keyboard at any moment. A task completes only through review or approval, with the evidence contract set by its task type — the API has no path to write the state directly — and the control plane stores no credential bodies: it knows paths and references, nothing more. Self-hosted on your infrastructure, with your own provider accounts.
- Separate pools — accounts, models, hosts, projects, roles — matched only at run time, on the record
- Immutable route decisions: score components, penalties, and every elimination with its reason
- Real execution: the providers’ own CLIs in tmux sessions that survive browser, deploy and SSH drops
Diloop
Diloop rebuilds language practice around the moment that matters: saying a sentence of your own. You start the microphone yourself — or type — in a real situation like ordering at a café; your recording becomes text and nothing is sent until you confirm it; the reply comes back in your learning language, with audio; and the correction is explained in your interface language, so you chase your next attempt, not a score. Around that loop sit lessons, scenarios, themed tracks and games, level discovery that suggests a starting point, and progress you can see — results, attempts, personal bests. Nine languages, with interface and learning language fully independent. It is in open beta and honest about it: replies and feedback are AI-generated, level discovery is not a certification, and beta content is still growing.
- A speaking loop you control: say it, confirm the transcript, hear the reply, see why
- Corrections explained in your interface language — next attempt over score
- Lessons, scenarios, themed tracks and games in one learning experience
Haico
Haico collapses the fragmented night — an event in one app, a ticket from another, the plan in group chat, the memories lost in a gallery — into one sentence. Chat is the interface: no menus, no forms; describe your intent, and Haico builds a plan card, shows what it understood, asks only for what is missing and leaves the final word with you. Payment, invitations and publishing are approval-gated — nothing moves until you confirm. After the door, Live Mode opens: features that exist only inside that event — who is here (friends-only by default), song requests to the DJ, live polls, and moments that wait for tag approval. Organisers run the other side the same way: describe the event in a sentence, and Haico drafts it, builds the checklist, collects vendor quotes, staffs the door and tracks sales. And everything is measured at the door, not the click: named tickets with rotating QRs, verified check-ins, connector earnings released on arrival, and an Experience Passport that stamps every night you actually attended. Live in Istanbul, with real events on the board.
- Chat as the interface: one sentence in, a plan card out — with what Haico understood, shown
- Approval-gated agency: payment, invites and publishing never happen without your confirmation
- Live Mode after check-in: who is here, DJ requests, live polls, consent-gated moments — per event
TRUST & SOVEREIGNTY
Your data never leaves your orbit.
Registered in the Netherlands, deployable on-prem or fully isolated, engineered for GDPR and KVKK. Sovereignty is a setting, not an upsell.
Netherlands-registered
On-prem or isolated on request
GDPR & KVKK by design
Build on the platform beneath the products.
Bring your workload to a sovereign stack you fully control.