Pressarch-a decentralized and privacy-oriented search engine-has just launched pregpt 2.0, which marks the newest attempt from the company to challenge the dominance of Big Tech in the AI Chatbot room.
The new release brought improved language models and a wider selection of open-source AI options, all performed on a network of distributed computers instead of centralized data centers.
“Why am I so excited? Because pregpt 2.0 is so powerful and unrestrained that it has the potential to fundamentally disturb the echo chamber effect that has long manipulated conventional wisdom, thereby reinforcing herd instinct in blind conformity, ”said Brenden Tacon, Innovation and Operations for the Research Leiden, told Decrypt.
The improved chatbot comes with two subscription layers: a monthly basic plan of $ 2 with the 7B model from Mistral AI and a $ 5 pro version powered by Venice.ai’s more refined LLMS. Both options promise to keep user data private and not -controlled conversations, with chats permanently deleted when deleting.
Model-Line-Up from Pregpt 2.0 contains six of the most famous names in the open-source AI room: Meta’s Llama-3.1-405B (a gigantic model), Lama-3.2-3B (a very small model built for efficiency) and Llama-3.3-70B (the newest LLM), Alibaba’s Qwen 32b.
It even uses the old Dolphin 2.9 model, previously known in AI circles because it is completely uncensored and powerful -and good in role play. The company also seems to have refined the Mistral 7B model to offer an adapted version.
“This model gracefully deals with a context of 8,000 tokens, which corresponds to around 5,000 words, and you will continue to 1000 messages per month,” said the company’s website.
This means that the model will have a memory of 5000 words and is unable to process conversations well that exceed such a limit – or will not process prompts that are so long.
What is PreSearch?
Presearch, which was launched in Beta in 2017 and went live in 2018, is actually a project that wants to restore search engine architecture with decentralized technology.
The platform processed more than 12 million monthly searches via a web of independent nodes. Each junction operator turned off pre-tokens and borrowed computer power to the network, creating a self-sufficient ecosystem that has been scaled naturally with the demand.
The idea is that a decentralized network makes the profiling of users – the business model of Google – and can help to generate a business model that is more transparent and more organic.
The advertising model of the platform also differs from what you see in Google or Bing, for example.
Instead of offering wars for keywords, advertisers raised pre -tokens to get visibility. The more tokens they use, the better their placement will be – a system that reduces token circulation and at the same time made predictable income.
Part of these tokens is burned periodically and gradually lowers the total supply of the current 590 million plus in circulation.
Pregpt 2.0 used this distributed infrastructure by working with Venice.ai, a privacy-conscious AI-service provider, and Salad.com, a community that shares decentralized GPU power.
The professional layer works on the powerful network of Venice.ai, while the basic plan is supported by the distributed GPU network of Salad.com.
Both paths encrypt user interactions and remembering the storage of chat logs, whereby the dedication of presess for privacy is maintained.
The tokenomics of pre keeps the system smooth. Users earn a maximum of 8 tokens every day for searches, while node operators receive rewards based on their ring size and search volume.
This seems, at least in theory, a win-win situation in which both users and advertisers are well rewarded and at the same time helping the ecosystem grow.
Pregpt 2.0 is a separate AI function that has been added to the Voorearch toolkit; The company remains aimed at its core mission of decentralized, private search assignment.
The chatbot integration is intended as a supplement to the seeker experience without overshadowing it.
The goal is to make the entire platform ideal for privacy-conscious users who want a replacement for traditional searches on the web and are curious about using AI tools in their daily lives.
Hands-on with pregpt 2.0: promise and limitations
Testing Pregpt 2.0 revealed a capable chatbot that prioritizes the function above flash. The interface felt cleaner than competitors such as Venice.ai or hugging chat, although it was missing the possibilities for generating image generation that have become standard elsewhere.
By integrating a system prompt function, users can refine the behavior of the AI through custom instructions, which is useful for getting more precise answers-a sound system prompt can dramatically increase the performance of a model.
The overall experience will be familiar with those who are used to tinker with different chatbots.
This was not a revolutionary jump in AI possibilities, but rather a privacy-oriented implementation of existing open-source models that are often less powerful than regular alternatives such as GPT-4O or Deepseek.
The platform only manages ordinary text. It can make a story before bedtime or summarize trends, but the lack of support for Excel documents and CSV files, PDFs or third-party documents cannot make it up correctly.
Instead, users must actually copy and paste the content of a leaf, which is far from ideal. Those who confuse decentralization with low speeds have nothing to worry about.
The answers were fast and the chatbot never hung. But the models offered the quality that you would expect from Open-Source LLMS that are not really the charts in the LLM arena at the top of the top LLAMA 3.1 405B is currently in the 27th position and is the most powerful model in the selection of the Searoar.
It is not bad, but it is also not impressive according to today’s standards.
There are currently some open-source implementations that are much more powerful in demonstrably comparable sizes.
Llama-3.1-Memotron-70B instruct can, for example, easily replace the newer (but not better) Llama-3.3-70B, and Deepseek R1 is jumping for the Llama 3.1 405b from Meta, the best open-source model to date.
In general, the experience was pleasant; The models that were implemented as expected, and the interface was easier to use than Venice AI, the most important competitor.
If you are looking for a privacy solution or want to try any AI tool that is available today, this function is definitely worth a look. Keep in mind that the search engine will not replace Google, and the AI chatbot will not replace chatgpt – at least not yet.
Published by Josh Quitittner and Sebastian Sinclair
Generally intelligent Newsletter
A weekly AI trip told by Gen, a generative AI model.
