An AI colleague that is honestgroundedfluentyourslive in 10 minutes
Reads your website. Replies in your tone. Cites every answer. Escalates only when it can't help. No prompt-engineering PhD required.
From website URL to a working AI colleague in ten minutes.
Six steps, most of them automatic. Then it gets better every week.
Paste your URL
Keloa reads your homepage, FAQ, product pages, and shipping policy. Auto-detects brand colours, logo, language, and tone of voice.
Try, tweak, refine
Sandbox the AI with real questions. Adjust tone (friendly / formal / playful), add phrases to use or avoid, and edit the system prompt directly when you want full control.
Q&A pairs in one click
Ask a question the AI couldn't answer? Add the answer once and it's stored as a Q&A pair, vectorised and live within seconds, no re-crawl required.
Drop in widget or email
Paste the chat snippet on your site, or forward your support inbox to your Keloa email address. Live in under ten minutes.
Cite count + escalation reasons
Per-page citation counts show which knowledge is pulling weight and which is dead ballast. Every escalation comes with a reason so you know what's missing.
Auto-recrawl + gap detection
On Business+, Keloa re-reads your site on schedule and surfaces unanswered questions on your dashboard. The AI gets better while you sleep.
Every answer carries its receipts.
Keloa cites the exact source behind every reply, the article, the product page, the order number. When the source doesn't cover the question, it says so and escalates. No hallucinations. No made-up policies.
- Inline citations visible to your operators
- Confidence score on every reply
- Full audit log of every AI decision, exportable on demand
AI that's safe to put in front of customers.
PII redaction
Emails, phone numbers, IBANs and card numbers are stripped before any request leaves the EU. Nothing sensitive touches the model.
Topic guardrails
Define what the AI must refuse (medical advice, legal opinions, competitor mentions) directly in the system prompt. Keloa will politely decline and escalate.
Human in the loop
Run in suggest-only mode indefinitely. Graduate one intent at a time to full automation as confidence grows.
Full audit log
Every prompt, every source, every confidence score, logged, searchable, exportable. For your DPO, your auditor, and your peace of mind.
Multilingual replies
Any language your customers write in, 100+ supported. The AI detects the language on the first message and replies in the same language for the rest of the conversation. Dutch and English are most polished, with German, French, Spanish and the rest right behind.
Smart handoff
Below 70% confidence, below threshold on sentiment, or whenever the customer asks, a human takes over with a one-paragraph summary.
See your AI colleague answer your customers' questions.
Bring your URL. Free Starter, no credit card. Live in ten minutes.
Start for free →Common questions about Keloa AI.
Which model powers Keloa AI?
Keloa is model-agnostic and routes between leading frontier models, selected per task by quality and cost. Every request is logged and reproducible. Enterprise customers on Scale can pin a specific model class or bring their own model key.
Where is my data processed?
All storage and retrieval happens in our EU region (Amsterdam). LLM requests are routed to EU endpoints. PII is redacted before any model request. Full details in our DPA.
Can I use Keloa AI without replacing my helpdesk?
Yes. Keloa can run as a “second opinion” next to Zendesk, Intercom, or Gorgias, or replace them entirely. Most customers start side-by-side and consolidate within 90 days.
What happens when the AI doesn't know the answer?
It says so, summarises the customer's question, tags the conversation with the missing topic, and routes to a human. Over time you see the “unanswered topics” report, a to-do list for your knowledge base.
How do you prevent hallucinations?
Retrieval-augmented generation with strict grounding. If the retrieved content doesn't cover the question above a confidence threshold, the AI does not generate an answer, it escalates.
Do you train on our data?
No. Your conversations are never used to train shared models. You own your data, we process it. Export or delete everything in one click.
Put your AI colleague to work today.
Starter is free forever. Setup in under ten minutes. We'll even migrate your knowledge base from Zendesk or Notion for you.