AI assistant platform
Kyros turns your documents, URLs and FAQs into an AI assistant grounded in your content — embed it as a website widget, integrate it into your own apps, or call it via API. GDPR-compliant, EU-hosted.
Of course — order #4021 hasn’t shipped yet, so I can move delivery to Saturday, 12 July. Want me to switch it?
Built for the DACH region
The AI chatbot & video-bot platform for teams who take compliance seriously.
By omniratio UG · built in Germany
Build once, deploy anywhere
The same configured assistant — same model, same knowledge — deploys as an embeddable website widget, inside your own product, or through a simple API. One assistant, every channel.
Website widget
Fully brand-themeable — colours, fonts, logo, launcher and position — with a copy-paste script or iframe. Works on any site or CMS.
Apps & API
Query your assistant programmatically over a simple API, and let interactive Cards call your own APIs or run your app’s code — so it doesn’t just answer, it acts.
Video avatar · optional
For sites that want a face: a lip-syncing avatar with gestures, in German (incl. dialects) and English — the same assistant, on screen.
One brain, every channel — what you tune in Studio ships the same everywhere: as text, in your app, or on screen.
From name to live
No template gallery, no runtime decision, no setup wizard — Kyros wires up the plumbing so you go straight to the work that matters.
Creating an assistant asks for exactly one thing — a name. Submit, and Kyros silently provisions a dedicated Knowledge Base and a ready-to-configure chat widget, then drops you into Studio.
Crawl URLs in bulk — including direct links to hosted PDF, Word or Excel files — or hand-write Golden answers the assistant returns verbatim on a matching question.
Pick a model, write the instructions, turn on content moderation, and test against a live preview. Save Variations, sandbox edits in the Playground, and go live only when you click Apply to assistant.
Publish as a chat widget and/or video bot — brand-theme it, set domain allow-listing and a GDPR privacy panel, then copy the embed snippet into any website.
Replay every real conversation in Chatlogs. Catch a bad answer, click Revise, and the fix writes back into the Knowledge Base as a permanent Golden answer.
What's under the hood
The features that separate 'edit and pray' from 'draft safely, go live on purpose.'
Save named snapshots as Variations, test unsaved edits in the sandboxed Playground, and push to real users only via Apply to assistant — with a computed 'Live' badge that never lies about what's running.
Run two full live configurations in parallel — each with its own model, reasoning effort and per-message credit cost — and fire one prompt at both with Send to both.
Answers cite their sources with clickable footnotes, narrate their progress ('Searching knowledge base…'), and admit when they don't know instead of guessing.
Hand-write exact Q&A pairs for the questions that matter most — a match returns your approved answer word-for-word, bypassing document retrieval.
In Chatlogs, turn a wrong real-world answer into a permanent Knowledge Base correction in one step — click Revise, edit, save.
Design multi-screen in-chat mini-apps — lead-capture forms, product tiles, approval flows — by describing them to a built-in AI editor. Then wire real actions: call your own API, run custom app code, jump between screens, or send a visitor a form to submit right in the chat.
A Knowledge Base is a standalone, reusable asset — write a policy set once and reuse it across every assistant that needs it.
Screen every incoming message for PII, sexual content, hate, violence, self-harm and more — up to ten categories, six on by default. Flagged messages are withheld before they reach the model or get stored, with a polite on-brand refusal.
Who it's for
From a solo founder to an agency juggling a dozen clients — Kyros's workspace model and permissions bend to how you actually work.
Put a grounded assistant on your site, then use Chatlogs to spot-check quality, read visitor ratings, and Revise weak answers into permanent fixes.
Brand-theme the chat widget and video bot to match your site exactly, and hand visitors interactive Cards instead of a wall of text — no code required.
One login, unlimited separate Workspaces — each client's assistants, knowledge, teammates and usage stay fully siloed, and you switch between them in two clicks.
Go from a single field to a live assistant in under a minute, with transparent per-message pricing and a 14-day free trial — no engineering team required.
Nothing — assistants, knowledge, members or usage — ever crosses between workspaces.
Multi-provider, transparent
A multi-provider model router with transparent per-message credit pricing and automatic fallback — so cost is never a surprise and one provider outage never takes your assistant down.
Every model shows its exact credit cost per message before you commit — and each reasoning-effort level is priced individually.
Models outside your plan aren't hidden — they appear in the same picker labelled 'Not in your plan,' turning config into an in-context upgrade path.
A router with automatic fallback underneath keeps the assistant answering even if one AI provider has an outage.
Compare Models runs two configurations against live, independent conversations — so you decide on evidence, not vibes.
Compliance, built in — not bolted on
GDPR wording inside the widget, per-assistant data retention, a full audit log, and resource-by-resource permissions — because 'trust us' isn't a compliance strategy.
A GDPR-worded notice — data processed, legal basis, retention, processor relationship — ships inside the widget's own menu, with no extra build work.
Choose platform default, never delete, or a custom number of inactivity days before conversation data is permanently purged — active chats are never swept away.
Every meaningful workspace action is captured across 21 tracked action types — actor, timestamp, before/after diff and status — filterable by action and date.
Restrict which sites may embed the widget (with wildcard subdomains), and sign embed codes with an expiry so you can rotate or time-box them.
A base role plus reusable Custom Roles plus per-member overrides across every resource — with Added / Removed / Inherited labelling so a save never changes access invisibly.
Hosted in the EU and bilingual by default (German primary, English secondary) — built for the DACH market from the ground up.
Guardrails that screen for PII, sexual content, hate and more
Content moderation — screened categories, before generation
Audit log — every workspace action, filterable
Start now
Name your assistant and it's already wired to its own knowledge base and chat widget. No template picker, no credit card — just start.
Nothing goes live until you click Apply to assistant — draft safely, go live on purpose.