Private AI Reviews

My main use case for Private AI involves long-running jobs. If I need thorough research, I don't want to burn my cloud AI tokens. I have a second brain project that indexes Markdown notes, a couple thousand of them. Every night, Private AI goes into this vault, cross-references the notes between each other, cleans it up, and if I add YouTube links, it transcribes them and adds context.

I run a trading pipeline that analyzes every hour what is happening in the stock market for specific commodities. Since I run it 24/7, using cloud AI would be very expensive. That is why I use Private AI, mostly GLM 5.2, or if my main box is busy, I use smaller models with around 27 billion parameters.

I use Private AI often as a replacement for frontier models such as Anthropic Opus or Anthropic Claude.

The best features Private AI offers include that besides the cost of hardware, it is essentially free to run, and I can run it 24/7 without worrying about API token bills. I also value that open-source models are catching up to frontier models. I am very satisfied with GLM 5.2, and I use it in exchange with Anthropic Opus frequently, often getting better results from the open-source GLM model.

What makes the open-source models stand out for me compared to frontier models is that I could not afford running jobs 24/7 with proprietary solutions, which fundamentally changes my workflow.

Private AI has positively impacted my organization. Although we are not running any Private AI models ourselves, we are the provider for Private AI. Other companies run Private AI models on our hardware.

Regarding how Private AI can be improved, currently serving open weights models is complicated. There are many parameters that you have to adjust to make it run well. The ecosystem is maturing and becoming easier. Tools like Ollama allow running models with no friction, but with suboptimal performance. The speed of development is important as the ecosystem is advancing rapidly, and the tools and overlays need to catch up to make it easier for normal users.

I have several computers of different kinds, such as a Mac Studio, a box in my house with NVIDIA GeForce cards like 3090, and a rented server with H100s or H200s. I do not know which quantization I should choose because quantization is complicated. The performance of models is affected by quantization type, MTU, token predictions, KV cache type, and cache size. Recipes for running models better would be appreciated. Some providers like Ansloth and G-Lang provide them, but they cover so much hardware that it is really hard to make a choice.

I have been using Private AI for about two years.

Private AI is stable.

The scalability of Private AI is fairly good. You just throw more hardware at it. There are very good frameworks available for scaling it up, such as DLRM. Some organizations are using one model company-wide, so with enough hardware, it is very scalable.

I do not know about customer support for Private AI because I only use the open-source models, so there is no customer support there. I run them myself. If I use some providers such as OpenRouter, I did not need to ask the provider for customer support.

I have not used any other solution before choosing Private AI.

Before choosing Private AI, I evaluated other options. At the same time I have several subscriptions for closed models with Anthropic and OpenAI, and I use one for one task and another for other tasks. Generally, if I have time-sensitive work to do, I use closed model providers such as Anthropic. If it is not time-sensitive and I can let my server run overnight, which is slower with bigger models, I check the results the next day.

I have not seen a return on investment and cannot share any relevant metrics such as money saved, time saved, or fewer employees needed, either for my company or for my clients.

My advice to others looking into using Private AI is to list your use cases and determine if you really need a very intelligent model or if you can manage with a smaller, possibly less capable but reliable model because that affects the price of the hardware that you need to own or rent. Stating the use cases is the most important part, and then you can derive what model you really need.

Regarding Private AI's AI capabilities, governance and security are very important because we are not sending our data to someone in the cloud. We do not know if our data is used in those models, even though they claim they do not use it and do not log it. From a data privacy perspective, this is crucial. About sovereignty, OpenAI and Anthropic, especially Anthropic, are censoring the models significantly, which is very sensitive. The models refuse to answer even simple questions that are not harmful or dangerous. From a geopolitical point of view, I believe this is very important. Currently, China is providing the best open-source models. There is some development in the USA. Europe is lagging behind, but I hope it will change. Private AI is resistant to censorship and bans from certain countries on the use of models. If Trump were to invoke a ban on the use of Anthropic models for non-US citizens, which is basically happening with Anthropic's and similar models right now, with Private AI, we would be protected from political influences.

Regarding Private AI's AI capabilities, accuracy and reliability of output are getting better continuously. I already mentioned that GLM 5.2 is on par with Anthropic Opus, which is excellent. However, the main barrier is the hardware one owns or rents to run it because these are very large and expensive models. Meanwhile, smaller models with around 30 billion parameters easily fit into Mac Minis, so they can be used locally without connecting to the internet. Things are getting better, but the main barrier remains the cost of hardware to run these models. I would rate this review a seven overall.

We're working with the Jetson Orion 64 gig unit and we also have a DGX Spark 128 gig GPU unit to implement our own AI solutions for on-premises customers. Usually, it's organizations that have to keep their data private because they cannot put it into the cloud for compliance reasons. Legal is a big one, medical is another field. Using those edge devices, we're able to do the work that cloud providers would do but keep it in a data sovereign solution.

We're focused on the edge, buying devices that are all-in-one. I was always trying to do most of our workload that we could with our own devices. I've been experimenting with agentic use on our DGX Spark. It is using my Hermes agent to do research and specific programming, which serves as an internal tool within Kestral.

For us, it's been very helpful in allowing me to come up with custom one-pagers for each customer, how I can focus in to help with their organizations based off of what I know. That allows me to do those changes very quickly.

I think what needs to happen is more organizations need to get into the open-source part of it. More countries even need to get into open-source models and creating their own or using others for baselines to then create their own. It's not just chat LLMs that are in there. The open-source community is very good and has some very good solutions on everything needed for AI from transcriptions to image generation to video generation and all that other work.

The pricing part has just been quite frankly insane in the last year. The jump in how much compute is costing has been significant. However, since I'm on the edge and I buy devices that are already there with everything in the box, the hardware has been available even though pricing has gone up exponentially.

I did not evaluate other options. I was already pursuing Private AI because I knew from my old business of 20 years where I was working in enterprise CCTV implementations that data sovereignty was always going to be something organizations would need. I knew organizations were always going to need that scenario where they don't want data going into the cloud whatsoever. I knew that there was a market there in the corporate side and I pursued that.

I can do everything with Private AI. I can have it make my own videos, have it generate my own images, be my LLM, be the AI for my agentic tool calling and my agentic workforce. When I don't have to worry about token costs anymore, that's where the innovation happens. Everyone now is starting to be experts in managing token costs and that loses the innovation that AI allows me to get to.

I've been experimenting with agentic use on our DGX Spark. It is using my Hermes agent to do research and specific programming. I have it doing some programming for me, creating solutions that are used as an internal tool within Kestral. From my point of view, it's not as quick as using a foundation model in the cloud, but if I have it do something and it takes three hours at night to do it and I get up in the morning and it's done, to me it's just the same as if it was sitting there. One of the things about the agentic use that I think people really need to understand is that it's not a chatbot. If I'm using it, I tell it to go do something, and it's going to go off and it's going to be thinking about what it needs to do, plan, work with other agents in the planning process to make sure they've got a good plan to go and then execute it. Executing is writing files and doing tasks. All of that takes time, including installing things.

It has met my expectations. I don't expect it to ever be perfect. The way I've been looking at it, especially on the agentic side and playing with that, is I need to think of it as a junior employee where it's very smart at doing something but it just doesn't have the common sense or the knowledge of my organization to actually implement something without my help. I always look at it as I've got to manage and maintain it.

I would rate Private AI right now at about a six. Last year I would have rated it at a four and the year before I would have rated it at a two. It's increasing. The quality of what's coming out in Private AI or local AI is improving every time and it is really closing the moat between the foundation models. Now I'm able to run something locally that does a lot of the grunt work for what my organization needs and then I can rely on the cloud to do some research or help my AI plan things. I'm very much at that number.

It's a six now because that's where the models are at compared to what the foundation models are. I think it's going in the right direction to be a seven and an eight once new models start coming out that are much more improved. The thing that's really neat is that we are taking and coming out with 30 billion parameter models that are smarter and run more efficiently than 70 billion or 100 billion parameter models of last year. That shrinkage and that knowledge gained on a smaller model is very exciting because smaller models can be run on more edge devices. My overall review rating for Private AI is six.

My main use case for Private AI is that we have numerous intellectual property products, so we do not rely on public AIs like ChatGPT, Claude, or Gemini. We developed Private AI instead of a public AI using certain tools to secure our product details.

A specific example of how I use Private AI to secure our product details is that we use a local LLM. Through that, we internally maintain a knowledge base that every person who is accessing the local network can access. It is not visible to the outside world and is not connected to the internet. The local LLM is the only knowledge base we are maintaining, which everybody can access for their day-to-day work, and that knowledge base is increasing daily through certain development teams.

My teams primarily use Private AI for tech pack analysis, analyzing product details, assessing what is beneficial or detrimental, the specifications, and how we can efficiently execute these products. They upload it to a local AI system and obtain insights that are intellectual property-bound information. That is the main reason why we do not use public AI.

The best features Private AI offers include the ability to upload two or three PDFs and get insights, such as summaries or comparisons between both attachments. Maintaining Private AI requires significant hardware infrastructure, and while we currently do not possess many high-capacity servers, within our allocated server infrastructure, we can perform a certain limitation of analysis and maintain chat histories.

The feature most valuable to my team in our day-to-day work is the analysis of PDFs such as tech packs, which is the most common feature. Besides that, they refer to the knowledge base for questions, and when a problem arises, they reach out to Private AI. Private AI answers through its knowledge base. When new people, such as new recruiters, join, they also receive help from that AI tool because the knowledge base is rapidly growing daily, specifically as a domain-oriented knowledge base not available in public.

Private AI has positively impacted our organization because the main intention of forming Private AI is to protect our intellectual property data. It helps significantly in terms of data governance and protecting customer data. Using Private AI instead of public AI ensures that when accessing public AI, we cannot assure or guarantee that our data and customer data are secure from being accessed elsewhere.

Private AI can be improved in that maintaining it requires significant hardware infrastructure investments. If we have that investment, we can establish Private AI for each organization so that they can have their own, which is very useful compared to commonly available AI tools.

Regarding needed improvements on the software side, we are currently using Open UI, which is freely available, but there is also a paid version. If an open UI extended version is developed, it would be greatly appreciated.

I have been working in my current field for approximately ten years.

The accuracy and reliability of output from Private AI depend on the knowledge base that the team has built. We are rapidly monitoring and continuously improving the knowledge base so that the team can access it based on recent experiments. The knowledge base is key.

My advice to others looking into using Private AI is that we hosted it on-premises for securing our data, but if someone prefers a hybrid option, they can certainly pursue that. I would rate this solution as a seven out of ten.

I use Private AI for several use cases. The first use case is professional reply because every day I have to reply to lots of messages and emails. I use Private AI to reply to those messages or emails. The second use case is professional rewrite because sometimes I need to draft emails and I will prepare the drafts and then copy those drafts to Private AI to polish and finish the draft. The third use case is email drafting because sometimes I only copy the email to Private AI or tell it what email I would like to send, then let it draft the email. The fourth use case is translation because our company's language is English and French. Sometimes I have to use Mandarin, but there is no Mandarin translation provided by our company, so I use Private AI to translate English or French material into Mandarin Chinese. The last use case is the coding agent or technical assistant because I have to use many technical tools including Power BI, Tableau, SQL, Alteryx, and Power Platform. When I use those technical tools, I ask detailed questions to Private AI and let it guide me step by step.

For example, I am currently working on building a Power BI dashboard. However, I did not know how to change the data format from text to number in Power BI. I asked this question to Private AI, and it guided me step by step about how to change the data type.

I believe the best feature of Private AI is its security because our company does not allow us to use other public AI models. We cannot use OpenAI, ChatGPT, Google Gemini, Anthropic Cloud Code, or Codex tools. All of those public AI models are prohibited in our company. Because all those public tools are blocked, we only have one choice in our company, which is using our company's Private AI. As recommended by our company, this tool is secure and safe, so we have no concern about using it. That is the best feature of Private AI from my perspective.

Private AI has also improved efficiency considerably. We can save a lot of time and spare more time for more valuable tasks.

I hope Private AI can achieve better performance because each time I input questions, I have to wait approximately ten seconds or sometimes twenty seconds to get the answer. I hope it can improve the speed and overall performance so that I can get answers in one or two seconds instead.

The performance issue prevents Private AI from being a perfect ten because I have to wait for a long time before getting the answer.

I believe the suggestion is to improve the performance before deployment so that we do not have to wait for a long time when using Private AI.

I have been using Private AI for more than one year.

Private AI is stable.

I believe Private AI's scalability depends on the server we use. Since we use a private cloud server, the scalability is quite good.

Because Private AI is deployed by our company, the support is also very good.

Private AI is currently the only solution. We do not use any other different solutions.

I believe Private AI is currently quite simple. It is published on one website, so I just log in to that website and can start to use it. The feature is similar to a browser and is very simple.

I believe there is currently no return on investment.

I believe Private AI is charged based on the tokens we use. Currently, I believe the cost is quite high.

I have summarized all of those use cases in my job.

Because Private AI was introduced by our company, we have no concern when we use it. I can ask all of my questions without worrying about any data leakage or compliance issues. That is the best part.

Private AI has saved me an average of one hour per work day. The total daily working hour is seven and a half hours.

Because I am not at the manager level, I do not know the relationships with the vendor.

My overall review rating for Private AI is eight out of ten. The interview was good, but please ask fewer questions next time and be more simplified and precise.

The main use case of Private AI is that you can use the AI in your private laptop environment. Since we are working with very crucial data that we do not want to share publicly, we can use Private AI to use the AI ourselves only, not publicly. We restrict Private AI and use the AI with our data specifically, not beyond that.

I decided that Private AI is the right fit for handling confidential information because the whole data can be captured on your PC or data center, and the public and outside entities are not able to see or analyze your data. That is why Private AI is a main concern for this. When we are dealing with patient data and any confidential data which are strictly private to a person, users, or a company, we have to use Private AI to manage or work with that type of data.

When we are dealing with data that are very confidential to the user or company, public AI usage is not a good fit for such cases because the data is given to all of the world. That is the reason we are using Private AI because the company does not want to lose their data in a public domain. They want their data private only. In the medical sector also, patients do not want their data on a public domain because it can be a disaster for them. For some of the injuries that are private to a person, they do not want them to go to the public room. That is the use case that can be used in the private area.

Private AI has improved technical performance because we have a large data set in Databricks and we are able to make queries in different notebooks, but it can take a longer time to read the file and process it. However, in Databricks, we have FireSpark and other technologies by which we can read, modify, and analyze processing data in a faster way. For that, we have to learn PySpark and other technologies. So for that, we use Genie AI and Private AI to make queries from PeerSpot.

Private AI can do better and currently is very good. However, in the future, they can also simplify the integration, scalability, and security. By which we will be able to integrate it and secure it and the appliances are very easy.

The current security of Private AI is good, but we would also want in the future that it can be increased and more reliable.

The accuracy of Private AI is far better as compared to the current scenario. It is not bad, but in the future, it would be more accurate.

Customer support on Private AI is less useful as of now, but I want it should be great in the future.

I would like to give advice to others by looking into Private AI's tab. We can first learn how all of these technologies work in the background and then apply them on the dataset on the application.

These are the points from myself which I want to implement into Private AI, where Private AI is most useful to others also. I would rate this product an eight out of ten.

We install for this client a very large on-premise architecture for forecasting and analysis of time series. In particular, this time series was related to selling the time series of sales in their rental stores. We produce the forecasting of sales of a particular product in a particular store for all stores belonging to the retail chain. For example, each day we perform about one million forecasts, forecasting one million time series.

For legacy AI, such as classic machine learning or time series forecasting, it is easy to build on-premise architecture.

I have used Private AI since the beginning of my career. In particular, we work extensively with manufacturing industries, and for the manufacturing industry, the usage of on-premise architecture is a must-have. Since 10 years ago, I started creating architecture on-premise for running AI models. Currently, I also develop architecture for running large language models.

Private AI is useful for not exposing sensitive information. The main goal of our clients is not to expose sensitive information in the cloud.

Without Private AI, we cannot face some clients because some of our clients must have private AI architecture; otherwise, they will not introduce AI in their process because they have to protect their sensitive information. From our point of view, this permits us to acquire many clients that we could not acquire and could not work with without this capability. Currently, 50% of our clients are with us because we can provide them private AI solutions.

Regarding the increase of performance, it is not a matter of performance but rather a matter of creating on-premise architecture to prevent sensitive information leakage.

With Kubernetes technology, it is easy to scale horizontally. The bottleneck is the hardware provisioning. If you have a good partner that can provide the hardware and move fast on this, it should be a good point to start.

Compatibility is given by Kubernetes. Kubernetes is the standard de facto for creating this type of on-premise architecture. Kubernetes permits deployment of containerized solutions and microservices, and also, for example with VLM, we can deploy large language models. With this technology, we can face on-premise architecture.

The provisioning of hardware is unique because we are a software company. We identify and provide to the client the hardware, such as the GPU for creating a Kubernetes cluster and running large language models. Provisioning the hardware is the most difficult task. For this, we use partners, such as Lenovo.

The most important challenge is related to developing and providing generative AI solutions. For example, deploying a large language model of 70 billion parameters is not an easy task on an on-premise architecture because you have to provide many GPUs. For the deployment of big large language models, it is a challenge. For many clients, we try to deploy small language models specific for a particular task. Providing a fast way for creating an on-premise architecture for large language models should be an important challenge for the future.

I have used Private AI since the beginning of my career.

I think that it is not as stable as a cloud architecture. You have to invest a lot of money in order to maintain the same type of stability.

The bottleneck is the hardware provisioning. If you have a good partner that can provide the hardware and move fast on this, it should be a good point to start.

The customer support is bigger than a cloud architecture because the customer support is provided by the cloud provider for cloud AI. On Private AI, we have to provide customer support. We have a contract with the client in order to manage scalability, durability, and continuity of service.

I rate the customer support a nine.

First of all, we used cloud AI, such as services provided by AWS, Azure. When the client asks us to move to a private cloud or on-premise architecture, we developed that type of solutions.

Governance and security are the most important aspects. Security is important because they have to protect the sensitive information belonging to the client. We develop many software components in order to guarantee the security of the information and particularly the authorization part. For example, when we deploy agents and AI ecosystems of AI agents, one of the most important aspects is to guarantee the authorization, such as the right access and the right information for each agent. This is one of the main challenges.

Governance is one of the most important tasks for us versus the client. The client pays us for the governance of this process and in particular the automation of the governance of this process.

I am a vendor of Private AI solutions.

Regarding the return on investment, the acquisition of new clients that needed private AI is a clear benefit.

We sign application maintenance contracts with our clients. We have software licenses for our software components. For the management of the hardware, that is a deal with our partners.

A hybrid approach is an option when some types of services can go on Private AI and some other types of services, such as services that do not impact sensitive information, can remain on the cloud. A hybrid architecture is a viable option.

My first advice is to understand very thoroughly the process of the clients in order to identify which types of services must remain on a cloud AI approach or the types of services that can go to Private AI solution because they manage sensitive information. My overall rating for this product is eight.

Private AI is used to interact with our security team to manage the agentic AI for the security practice. We have developed a system with 36 different types of agents, and we use a private LLM model to manage the interaction with our customer's system.

The Sentinel agent manages the privacy of customer data, and using a private LLM model is necessary because it is not possible to use a public model such as ChatGPT or similar.

We use data from customer tickets to deep learn our model to specialize it in different types of activities, for example, SOC, NOC, GRC, and similar. Specializing the model for specific jobs helps our team as we have built a RAG system using company data, ticket data, and data from interactions with customers, such as chat discussions regarding tickets or translations of phone calls with agents. We use this data to deep learn our LLM model to specialize it in a very specific way to manage high-risk activities, such as restarting a system that was broken by an incident.

The best feature of Private AI is the possibility to specialize the model for a specific type of job.

Private AI has positively impacted our organization in terms of cost; we can speed up from 10 to 30 times the number of activities we can manage with the same human team. In terms of problem-solving velocity, we have sped up about 2,000 times before using this technology while also looking for specific types of activities to enhance operator growth.

We have reduced the cost per ticket from two Euros to 70 cents and multiplied by five times the number of tickets we can manage in the same days.

The improvement needs for Private AI are complex because the amount of money and time necessary to deeply learn the model is very high, but the difference between one month and one year of deep learning for our type of activity is not significant, which raises the question of how much time is necessary for training the model with specific data to achieve the best results in cost and performance growth.

Regarding Private AI's AI capabilities, governance and security pose significant challenges because, by the nature of the LLM, it is not feasible to address a problem while simultaneously ensuring the security of the resolution and the customer's data. Therefore, it is necessary to have different levels of agents and models that work together with rules to manage issues that may arise during activities.

The accuracy and reliability of output from Private AI depend on the maturity of the RAG system and the depth of training for the model. In relation to specific technological problems, it should perform well, although I have experienced less effectiveness in other cases.

I have been using Private AI for five years.

Private AI is stable.

Private AI's scalability is impressive. We are using NVIDIA machines and can scale this system nearly linearly with our requirements.

The customer support is great. I would rate the customer support a nine.

We built this solution ourselves without previously using a different one.

We see a return on investment with 800,000 Euros spent to build the system and an ROI of 3 million in the first six months of the year.

My experience regarding Private AI's pricing, setup cost, and licensing is neutral. Currently, I am not managing costs directly but focusing on technology division, capability building, and functionality development.

We did not evaluate other options before choosing Private AI as we operate in the security environment where understanding how the model works, its architecture, and interaction with it is central to our business, leaving no other viable options.

We plan to build a different type of agent that can be invoked for problems that arise independently from an event, such as monitoring a system and managing its issues autonomously, marking the next step towards developing a sort of human agentic AI.

My advice for others considering Private AI is to clearly define your needs as precisely as possible. The key point about Private AI is its capability to manage LLM technology without risks related to data safety and proprietary information. Additionally, it is crucial to build a vector database containing all company information, as a private model alone is inadequate without this integration.

I would rate this review a seven overall.

I have been using Private AI more actively over the last couple of years.

I primarily use Private AI as an assistant for research purposes to find different solutions and alternatives, as well as for debugging and sometimes running investigations. As an Architect, I am tasked with finding feasible solutions for enterprise systems to connect with each other or how to promote metadata from one environment to another. Private AI comes in handy, especially when dealing with very specific deployment errors that are not easy to understand. It helps me sort through a bunch of options and tools available in the market, narrowing them down to offer the precise solution to the client.

I have seen that Private AI helps not only in my role but also for my peers and the people in my team, as they are leveraging Private AI functionalities for business analyst cases or for developers as well. There are different types of use cases in our roles. The best features of Private AI, in my opinion, is its overall security because by using Private AI's private LLMs, we are ensuring that none of the confidential or sensitive data from the client is used to train anything. Everything is kept secured and not shared publicly. Security features of Private AI impact my day-to-day work and my confidence in using the tool because I work for the insurance industry, which is a regulated industry that needs to comply with many guidelines and rules, and one of the most important is security and privacy.

The best features of Private AI, in my opinion, is its overall security because by using Private AI's private LLMs, we are ensuring that none of the confidential or sensitive data from the client is used to train anything. Everything is kept secured and not shared publicly. Security features impact my day-to-day work and my confidence in using the tool because I work for the insurance industry, which is a regulated industry that needs to comply with many guidelines and rules, and one of the most important is security and privacy.

Private AI has helped us become more agile and speed up certain processes that used to be very time-consuming in the past so that we can focus on different topics that add more value to the business.

I think improvements for Private AI depend on the system; for example, if a client uses cloud computing or is stuck with on-premises servers. Private AI always needs a live connection, so there may be certain security rules to follow, such as firewall rules or using specific IP ranges, a VPN, or an intranet, depending on how secure the connection needs to be.

Regarding integration, improvements are probably needed to ensure it is well connected to most systems or, if there is no built-in connector, to give us the possibility to create a custom one via API, for example, API REST or some sort of API.

I have been using Private AI more actively over the last couple of years.

The scalability of Private AI depends on the plans and usage of tokens as well as how it is set up and used in the organization; there must be a certain type of governance for it to work as best as possible.

Most of the time, the vendor's responsiveness to my questions or issues is knowledgeable, but I understand if sometimes they do not have an answer for everything and can return later with one.

I would say customer support is adequate.

The company started using ChatGPT from OpenAI and then switched to Claude from Anthropic due to some security issues and changes in privacy policies with ChatGPT, which led the CEO to decide to move to a different provider.

For me as a data scientist, the onboarding and training process for Private AI was not a significant challenge, but I understand that other people from different backgrounds might have a hard time grasping it. For me, it is fairly easy to integrate Private AI with our existing systems.

When debugging, we were using a tool for document generation that was not user-friendly regarding error handling. It was like looking for a needle in a haystack whenever we encountered an error. Private AI tools were intelligent enough to look over the code and give us context to understand what went wrong so that we could fix it faster.

I have seen the benefit of keeping data privacy and security apart from just giving it to the public or opening it up to everyone else, but I would like to keep assessing more functionalities and their impact on other use cases we have not used yet. Private AI's accuracy and reliability of output have been pretty good so far; it is not perfect, of course, which is why it is important to have a human in the loop. When Private AI hallucinates or gives something that seems off, I always double-check and verify; I never take everything Private AI gives us for granted and am very critical and analytical regarding the answers.

My advice for others looking into using Private AI is to assess the pros and cons of what they are buying, ensuring it suits the needs of the business, and to choose a specific plan based on the number of users and how big the organization is. I would rate this review an eight out of ten.

Our main use case for Private AI is deploying our own model in our private cloud within our data center only, and we use it for inference purposes. We have CPU, GPUs, and all necessary infrastructure on those servers.

A specific example of how we use Private AI for inference is that we deploy and maintain it as a Docker container. We store our model in a registry, and then we call it through an API call. The model responds back, and that is how we perform batching and VLLM API inference. All of this is deployed in our cloud only.

Regarding our main use case setup and process, we have our own hypervisors, and based on the hypervisor, there are storage devices, physical devices, and virtual VMs. The VM gets spun up, and network devices are separately built where network connectivity is established. We have Kubernetes deployed in our VMs, and from there we deploy our applications through Jenkins. Rancher is the platform where we manage our applications, which provides a solid enterprise architecture.

The best features Private AI offers is that we have full control, and it is private and not affected by anyone.

When I mention full control and privacy, it benefits my team day-to-day by ensuring that our data is secure only because it is private. Our data will not get shared in public storage and will be stored in our cloud only. This is especially valuable for sensitive data that will not get disclosed.

Private AI has positively impacted our organization because our LLM jobs are running and we are conducting one lakh of hits every day. We are analyzing our own data and bringing it into business use cases and generating business insights from the data.

The business insights we have been able to generate from this volume of hits benefit our organization's goals by helping us understand the root cause on ITSM data. We identify what is happening, which customer is falling, what the pain areas of the customer are, what the improvement areas of the customer are, and what the main concerns about the customer are. Whenever we go into meetings, the business team should be aware of these insights. In terms of performance, use case, and optimization, management can see on the dashboard which particular customers need more attention and what needs to be done to reduce that. The recommended actions are generated within minutes.

Private AI can be improved by providing a higher number of token limits.

Regarding needed improvements, I feel that documentation, support, and deployment could be better.

I have been working in my current field for six years and have been using Private AI for six years.

Private AI is stable, but I am not certain how many people use it.

I am not certain about Private AI's scalability.

I have never been in touch with customer support.

I previously used open-source models deployed on our GPUs.

I have seen a return on investment from Private AI because time definitely got saved. We achieve tasks that we are supposed to do in a week within a day. Regarding money, I cannot guarantee it because the more tokens we use, the more expensive it may become. AI is not free and it is expensive.

Regarding my experience with pricing, setup cost, and licensing, the billing team is aware of that, but I am not certain about the specific details.

Before choosing Private AI, I evaluated other options including VLLM inference and then deploying models such as Ollama and Llama 3, which we have self-deployed.

Regarding features and performance, I have seen that it is quite good. It depends on which GPU you are running it on. Based on the L40, it does not support that much, but H100 does very well.

Regarding Private AI's AI capabilities, I believe what they have stated in the documentation is correct. I do not know if they breach their documentation or the agreement, so I cannot say.

The accuracy and reliability of output from Private AI are good.

The advice I would give to others looking into using Private AI is to consider the number of tokens that it provides per session.

My main use case for Private AI is for private data and coding.

A specific example of how I use Private AI is analyzing data from the server.

I do not have anything else to add about my main use case with Private AI, as that remains confidential.

Regarding Private AI's AI capabilities, the security is strong because it is a private AI.

Regarding Private AI's AI capabilities, the accuracy is respectable and the reliability has been excellent so far.

I do not consider any improvements needed for Private AI.

I have been working in my current field or industry for at least one year.

My advice for others looking into using Private AI is to use it for research, so Private AI can advise others because that is now private. I would rate this product a 7 out of 10.