Cikisi VS Perplexity

Now that the market understands how our search engine differs from Google, many organizations are asking us how Cikisi’s AI agent, Ask Mila, differs from Perplexity.

Valéry Mainjot, CEO of Cikisi responds:

« As CFO, I would say that our tech startup Cikisi has raised 1,200 times less funding than Perplexity over the last 12 months, and that it certainly generates less interest in a European venture-capital market that is often cautious and very little visionary!

As CEO, I would point out that Cikisi remains a tool dedicated to business information and OSINT professionals, unlike Perplexity. If in the course of your work, you have specific questions that you want to receive answers you can trust (without having to reread all the references), and all that confidentialy, then I invite you to use Cikisi! The content generated by our chatbot “Ask Mila” is free of hallucinations and is based solely on information from our Web index.

As we say here when we don’t know something: “no bullshit, no bluffing”! »

Major issues

❌ As you will have understood, the primary concern is data confidentiality and security.

It is very dangerous for European companies to trust public chatbots such as Perplexity. There are many known concerns associated with these artificial intelligences: collection of sensitive data or without explicit consent, use of these data for model training, and sometimes, information leak. These are just a few examples among many others, as IBM explains.

✅ With Ask Mila, you keep control!

Developed entirely by Cikisi developers, this module is based on proprietary algorithms and the latest and most powerful open-source technologies. Fully self-hosted, it does not depend on any third-party services, so your data are NEVER shared. And your data remains strictly confidential: they are NEVER used or integrated into the model’s learning process.

Our AI is trained exclusively on data from our Web index, but NEVER with our customers’ data.


❌ Secondly, it is unthinkable to let a tool like Perplexity provide you with erroneous information on which you could base your strategic decisions.

We believe that to operate intelligently, organizations should not settle for “roughly true” and certainly not for “not totally false”. Publicly available generative AI tools make mistakes.

• They handle Web searches poorly, meaning they don’t know when to perform or not a web search to supplement their knowledge.
  
  
• They generate hallucinations or approximations, compensating for their gaps with unverified invented information.
  
  
• They analyze too few sources, sometimes deliberately, in order to reduce research and processing costs and the waiting time before starting to write the response.
  
  A test was shared earlier this year. The same question had been asked to Ask Mila and Perplexity, this time adding ChatGPT and Copilot.
  
  The result? The accuracy and comprehensiveness of the answers were completely different:
  ▪️ Ask Mila provided 39 citations from 28 different sources;
  ▪️ ChatGPT provided 17 citations from 9 different sources;
  ▪️ Perplexity provided 7 citations from 7 different sources;
  ▪️ Copilot provided 3 citations from 3 different sources.
  
  
  Find out more: 👉 Click here
  
  

✅ Here too, with Ask Mila, there is no need to worry.

• The AI assistant ALWAYS searches before responding, using Cikisi’s Web index or data chosen by the user. In the first case, Mila uses complex queries that are not supported by traditional search engines.
  
  
• By default, Mila does not use the training data from the selected model to respond, preferring to refrain from responding rather than hallucinating or approximating.
  
  
• And it then analyzes up to 100 search results to provide you with a reliable and sourced answer.
  
  

Ask Mila, how is this possible?

We attach particular importance to Cikisi’s DNA: remaining in control of its data.

Ask Mila is a unique AI agent:

• 0 web searches performed outside our infrastructure hosted in France;
• 0 use of a third-party web service;
• 0 use of the general knowledge of the “open-source” LLMs we host;
• 100% our RAG on our index.
  
  

Concrete test

The same question had been asked to Perplexity and Ask Mila: “List security incidents in North Kivu for the last 7 days”.

❓ A simple query that includes a thematic context (security incidents), a geographic context (North Kivu) and a temporal context (last 7 days).

However, the presentation below shows that even though Perplexity’s response is very well written and appears to be correct, it is in fact 95% false and represents a real danger to our businesses. In stark contrast, Cikisi’s response is comprehensive and correct.

We have tons of examples like this at Cikisi.

Still not convinced? Request a demo and see it yourself: