On-Premise AI Agent Platform: Train Your Teams and Assist Them on the Job
By
Convai Team
November 6, 2026
Every team has the person everyone asks. Most of what they get asked repeats: the same expert answers the same questions 90% of the time. The worker waits, the expert switches context, and productivity is lost.
Training alone doesn't close that gap. Formal training is a small share of how people learn, probably less than 10%. Most learning happens on the job: at the machine, inside the application, in the middle of a hard conversation.
The answers usually exist already, in CAD files, manuals, procedures and training videos that nobody can search with their hands full. And in aerospace, defense or manufacturing, that knowledge often has to stay inside your own network.
Convai is one agent platform for both jobs: training your people and helping them while they work. Build an agent on your own servers or in your own cloud, connect it to your documents, CAD models and systems, and deploy it into your applications, your 3D simulations and the devices your people carry. The same agent and the same knowledge can then support someone learning a procedure in 3D and doing it on the real equipment.
The Convai platform in one picture
Before you evaluate any single product, look at what sits underneath it. Read the platform as a stack: your AI data center at the bottom, then the AI models and agent infrastructure, then the SDKs and products that bring agents to your people.
Full package: everything in AI only, plus pixel-streaming infrastructure and rendering engines, so 3D experiences such as avatars and Convai Sim stream from your servers to your users' browsers.AI only: the agent platform and AI models run in your data center, and your own apps and 3D engines handle rendering.
Agents in your web apps(served by Web SDK)Add an agent to your web applications and portals. It can interview a user and fill in a long form for them, highlight parts of the interface, and guide people through each task.
Agents in your 3D simulations(served by 3D engine SDKs)Add conversational agents to the simulations your team already builds in Unity, Unreal Engine, NVIDIA Omniverse or three.js, so trainees can ask questions and get guidance inside the scenario.
Trade-show and shop-floor kiosks(served by Avatar Studio)Put a conversational avatar on a kiosk at your trade-show booth, exhibition or shop floor. With a camera enabled, it sees visitors, answers their questions and brings up relevant visuals.
Soft-skills training(served by Avatar Studio)Let your people practice sales conversations, HR situations and leadership coaching with an avatar, rehearse difficult conversations, and score each session against criteria you define.
Technical training from CAD models(served by Assembly Studio)Bring in your CAD files and let people ask about parts and assemblies by voice. Generate step-by-step training procedures from the model for your experts to review.
Digital twins with AI guides(served by Convai Sim)Capture a site with Gaussian splatting, bring it into an explorable 3D environment connected to your CAD knowledge, and add conversational avatars that give virtual tours and act out procedures inside it.
Assistance with physical work(served by Vision Assistant)With the Vision Assistant on a phone or tablet, or an XR app built with our plugins, the AI sees what your technician sees and guides each step, using your manuals and documents.
Your applications and products
Web SDKEmbed an agent in your web apps, portals and pages. It can fill in forms through conversation, highlight parts of the interface and guide users.
3D engine SDKsPlugins for Unity, Unreal Engine and NVIDIA Omniverse, plus three.js on the web, add conversational agents to the 3D simulations you already build. The Unity and Unreal Engine SDKs also cover desktop applications.
Access you controlAgents reach your systems only through MCP servers and credentials you approve. Your API key stays on your server; browsers use tokens you issue. Connect Convai to your directory, such as Active Directory, so access to the platform follows the groups you already manage.
Avatar StudioBuild a conversational avatar once and use it on a kiosk at your booth or shop floor, or as a practice partner for soft-skills training.
Assembly StudioBring in your CAD model. Assembly Studio breaks it into parts, reasons about how they fit together, and drafts step-by-step procedures for your experts to review.
Convai SimBring Gaussian-splat captures and digital twins into an explorable 3D world connected to your CAD knowledge, with avatars that guide tours and demonstrate procedures.
Vision AssistantOn a phone or tablet, or in an XR app built with our plugins, the AI sees what your technician sees and guides them step by step.
One core agentThe agents you build are available across Avatar Studio, Assembly Studio and Convai Sim, and in your own apps through our SDKs.
Foundation: agent infrastructure
FoundationThe agent platform and AI models that every product and SDK above is built on, running in your data center.
Agent creationCreate an agent, configure how it works, connect your data sources, and deploy it into your applications, browsers and 3D environments.
Agent runtimeRuns your agents' live voice and text conversations in real time, inside your apps, simulations and devices.
ScalingScale your agents across your own GPU servers as more teams and users come on board.
ObservabilityOversee how your agents perform and how people use them, with centralized supervision for the teams that maintain them.
AuthenticationControl who can build and manage your agents. Your API key stays on your server, and browsers connect with tokens you issue.
Agent management(built in)Create, configure and update all your agents in one place, from their knowledge and voice to the models they use.
MCP connections(built in)Connect agents to your internal tools and data through MCP servers, using OAuth to authorize each connection.
Knowledge Bank(built in)Add documents, videos and other files. The platform processes them into multimodal knowledge that agents draw on in conversation.
Long-term memory(built in)With long-term memory, an agent remembers each user's preferences and facts from earlier sessions.
Mindview(built in)Mindview shows the full prompt behind any response, so your team can trace and fix unexpected answers.
Team workspaces(built in)Give each team its own workspace or share one, so groups can build, maintain and monitor agents together under central oversight.
Matchmaking servers(Full package)Assign each person who opens a 3D experience to an available rendering server on your hardware.
Signalling & proxy(Full package)Set up each streaming session and relay video and input between your rendering servers and the user's browser.
Asset manager(Full package)Keep the 3D environments, avatars and other assets your experiences use in one place, ready for your rendering servers.
Stream scaling(Full package)Add rendering capacity on your GPUs as more people open your 3D experiences.
Foundation: AI models and your data center
Speech to textTurns what your users say into text the agent can act on, so people can talk instead of typing.
Language & vision modelsLarge language and vision-language models let agents reason over what users say and what a camera shows.
Text to speechGives each agent its own voice, so answers are spoken aloud in your apps, simulations and kiosks.
AnimationNeuroSync, Convai's own animation model, turns an agent's speech into real-time facial animation and lip sync for avatars.
Multimodal embeddingsIndex your documents, images and videos so agents can find and use the right information during a conversation.
Bring your own modelsUse the models the platform provides, or connect the LLMs and AI services your organization already runs.
Rendering engines(Full package)3D experiences render on Unreal Engine, Unity or NVIDIA Omniverse on your own GPU servers, then stream to your users' browsers.
Your AI data centerThe platform runs as containers on your own GPU servers or in your own cloud, including air-gapped networks, so your data stays inside your network.
Add an agent to your web applications and portals. It can interview a user and fill in a long form for them, highlight parts of the interface, and guide people through each task.Add conversational agents to the simulations your team already builds in Unity, Unreal Engine, NVIDIA Omniverse or three.js, so trainees can ask questions and get guidance inside the scenario.Put a conversational avatar on a kiosk at your trade-show booth, exhibition or shop floor. With a camera enabled, it sees visitors, answers their questions and brings up relevant visuals.Let your people practice sales conversations, HR situations and leadership coaching with an avatar, rehearse difficult conversations, and score each session against criteria you define.Bring in your CAD files and let people ask about parts and assemblies by voice. Generate step-by-step training procedures from the model for your experts to review.Capture a site with Gaussian splatting, bring it into an explorable 3D environment connected to your CAD knowledge, and add conversational avatars that give virtual tours and act out procedures inside it.With the Vision Assistant on a phone or tablet, or an XR app built with our plugins, the AI sees what your technician sees and guides each step, using your manuals and documents.Embed an agent in your web apps, portals and pages. It can fill in forms through conversation, highlight parts of the interface and guide users.Plugins for Unity, Unreal Engine and NVIDIA Omniverse, plus three.js on the web, add conversational agents to the 3D simulations you already build. The Unity and Unreal Engine SDKs also cover desktop applications.Agents reach your systems only through MCP servers and credentials you approve. Your API key stays on your server; browsers use tokens you issue. Connect Convai to your directory, such as Active Directory, so access to the platform follows the groups you already manage.Build a conversational avatar once and use it on a kiosk at your booth or shop floor, or as a practice partner for soft-skills training.Bring in your CAD model. Assembly Studio breaks it into parts, reasons about how they fit together, and drafts step-by-step procedures for your experts to review.Bring Gaussian-splat captures and digital twins into an explorable 3D world connected to your CAD knowledge, with avatars that guide tours and demonstrate procedures.On a phone or tablet, or in an XR app built with our plugins, the AI sees what your technician sees and guides them step by step.The agents you build are available across Avatar Studio, Assembly Studio and Convai Sim, and in your own apps through our SDKs.The agent platform and AI models that every product and SDK above is built on, running in your data center.Create an agent, configure how it works, connect your data sources, and deploy it into your applications, browsers and 3D environments.Runs your agents' live voice and text conversations in real time, inside your apps, simulations and devices.Scale your agents across your own GPU servers as more teams and users come on board.Oversee how your agents perform and how people use them, with centralized supervision for the teams that maintain them.Control who can build and manage your agents. Your API key stays on your server, and browsers connect with tokens you issue.Create, configure and update all your agents in one place, from their knowledge and voice to the models they use.Connect agents to your internal tools and data through MCP servers, using OAuth to authorize each connection.Add documents, videos and other files. The platform processes them into multimodal knowledge that agents draw on in conversation.With long-term memory, an agent remembers each user's preferences and facts from earlier sessions.Mindview shows the full prompt behind any response, so your team can trace and fix unexpected answers.Give each team its own workspace or share one, so groups can build, maintain and monitor agents together under central oversight.Assign each person who opens a 3D experience to an available rendering server on your hardware.Set up each streaming session and relay video and input between your rendering servers and the user's browser.Keep the 3D environments, avatars and other assets your experiences use in one place, ready for your rendering servers.Add rendering capacity on your GPUs as more people open your 3D experiences.Turns what your users say into text the agent can act on, so people can talk instead of typing.Large language and vision-language models let agents reason over what users say and what a camera shows.Gives each agent its own voice, so answers are spoken aloud in your apps, simulations and kiosks.NeuroSync, Convai's own animation model, turns an agent's speech into real-time facial animation and lip sync for avatars.Index your documents, images and videos so agents can find and use the right information during a conversation.Use the models the platform provides, or connect the LLMs and AI services your organization already runs.3D experiences render on Unreal Engine, Unity or NVIDIA Omniverse on your own GPU servers, then stream to your users' browsers.The platform runs as containers on your own GPU servers or in your own cloud, including air-gapped networks, so your data stays inside your network.End of the platform diagramSelect any block to jump to its section, use the Map link above each section heading to come back, and switch between AI only and Full package to compare the two deployments.
Find the use case closest to yours and the part of the platform that serves it:
The foundation: an AI agent platform on your own servers
An AI that knows your equipment and procedures needs more than a chat window, and the foundation is where you build it. Create an agent, configure how it works, connect your data, and deploy it into your applications, browsers and other environments, then manage and monitor it from the same platform. Everything in the layers above builds on these agents, so the work you put into one agent carries across.
Your agent at the center, with your knowledge and systems on one side, the places you deploy it on the other, AI models underneath and governance on top.
AI models for speech, voice and vision, or bring your own
A technician asks a question out loud. The agent has to hear it, understand it, look at what they're showing and answer in a natural voice. The platform provides the models an agent needs: speech-to-text, voice generation, large language models, vision understanding, multimodal embeddings and long-term memory, plus NeuroSync, Convai's own model for real-time facial animation.
If your security team has signed off on a language model, you don't have to give it up. Use the models the platform provides, or connect the LLMs your enterprise already approves through an OpenAI-compatible endpoint. Convai also works with the AI services and infrastructure you already run, from agent SDKs and MCP servers to data sources and existing training workflows.
Agent infrastructure: Knowledge Bank, MCP and team workspaces
Create and configure an agent in one editor, with its Knowledge Bank, AI settings, MCP connections and memory beside a live preview.
Bring documents, PDFs, images, audio and video into the Knowledge Bank. The agent draws on all of it in context during a conversation, including what it sees and hears in your training videos. Upload a lathe training video, for example, and the agent answers setup questions from the footage.
To reach live systems, connect your agents to MCP servers. MCP (Model Context Protocol) is an open standard for connecting AI agents to tools and data. The agent discovers each server's tools and calls them mid-conversation to look up live data or take an action in your systems.
Around each agent, the platform runs the infrastructure it needs in production: the agent runtime, scaling, observability and authentication. Supervise every agent from one place: teams work in their own workspaces or share one, review past sessions, and monitor usage and analytics across your agents and the people using them. Mindview shows the full prompt behind any response, so your team can trace and fix unexpected answers.
On-premises or air-gapped, with access you control
Your IT and security teams will ask where the system runs, what it can reach and who controls access. Run the whole platform on your own servers or in your own cloud. Every component ships as a container, nothing leaves your network, and it can run fully air-gapped. Convai is SOC 2 compliant. Convai Enterprise lays out the deployment options.
Your agents reach your systems only through the MCP servers and credentials you approve, and those servers decide what they return. MCP connections can use OAuth: sign in to the provider once, and Convai stores the tokens encrypted and refreshes them automatically.
You also control who gets in. Your API key stays on your server, and the browser connects with a token you issue. Connect Convai to your directory, such as Active Directory, so access to the platform follows the groups you already manage.
A field engineer filing a long service report, or a new hire lost in an internal portal, needs help inside the application they're using. Add an agent to your web application or portal with the Web SDK. It can interview a user to fill in a long form, navigate the app, highlight what matters and bring up relevant information.
The agent asks a technician questions, fills in a field service report from the answers and highlights the next step.
Your application stays in charge. As the website assistant guide explains, the agent names an action, and your code decides whether and how to run it. With camera and screen share turned on, the agent sees what the user sees while it helps.
An avatar experience embedded in an existing web page, ready to help visitors.
Already building training simulations in a game engine? SDKs and plugins for Unity, Unreal Engine, NVIDIA Omniverse and three.js bring the same agents into your 3D simulations. The Unity and Unreal Engine SDKs also cover desktop applications, and they include MCP tools, so your coding agents can add, configure and troubleshoot Convai characters in your project.
Avatar Studio: AI avatar kiosks and soft-skills training
At a busy booth, or on a shop floor where nobody is free to answer, a face that listens does more than a printed sign. Put an avatar from Avatar Studio on a kiosk at a trade show, on the shop floor or in an exhibition.
With a camera enabled, it can see the people in front of it, answer their questions and bring up relevant visuals. You build the avatar experience in your browser, and it runs in a browser too, so the kiosk has nothing to install.
A visitor in a suit talks with an AI avatar on a large display at a trade-show booth
Two jobs for Avatar Studio: talking with visitors at a trade-show booth, and playing an upset airline passenger in a role-play.
A manager's first attempt at hard feedback, or a sales rep's first tough objection, shouldn't happen with a real employee or customer. Use the same avatar for soft-skills practice: sales conversations, HR scenarios, leadership coaching and difficult conversations. Then score each session against criteria you define with the Evaluation API.
Assembly Studio: technical training from your CAD models
For a new machine, the most complete description is often the CAD model, sitting in software your technicians never open. Bring that model into Assembly Studio, even a complex assembly. It breaks the model into parts, reasons about how they fit together in 3D, and links them to your manuals, documents and videos, so you can ask, "How do I use this device?"
It reads the mechanics, too. From the CAD file alone, Assembly Studio works out how each component attaches and the order in which parts come off, then shows every step with an animation built for your question, much like 3D work instructions.
Sample model: an industrial robot arm in exploded view, with the AI guide answering questions about its parts.
You don't have to build a course by hand, either. Describe a procedure in a prompt and Assembly Studio drafts a step-by-step course. Your expert reviews it, then the AI guides each learner through the steps and quizzes them as they go.
From CAD model to guided procedure, with your expert reviewing the course and the equipment knowledge joining your agent's knowledge.
Assembly Studio is also how CAD knowledge enters the platform. Once a model is ingested, it becomes part of your agent's knowledge, alongside your Knowledge Bank and MCP connections.
Vision Assistant: AI-guided procedures on real equipment
Learning a procedure is half the job; the other half happens at the machine, often with both hands busy. With Convai's Vision Assistant on a phone or tablet, or an XR app built with our plugins, the AI sees what your technician sees and guides them step by step through the procedure on the real equipment. Or set up a camera beside the workstation, so the AI watches the task like an expert standing next to your technician and helps them through it.
An expert beside the person doing the work: on a phone or tablet, through a camera at the workstation, or in XR apps built with our plugins.
On the job, the same AI knows your CAD files and connected knowledge sources, and answers at the point of work, wherever your technician asks. That is why Assembly Studio and the Vision Assistant work hand in hand: your people learn the procedure on the CAD model, then get guided through it on the real machine.
No CAD file? The Vision Assistant also works from the videos, PDFs, manuals and documentation you've already connected.
Some things are best learned in a replica first: a facility new hires haven't walked yet, or a procedure too risky to practice on the live line. Capture your factory floor, a machine or a component with Gaussian splatting, and bring it into Convai Sim as a realistically rendered 3D environment: a digital twin your people can explore. Then connect it with your CAD knowledge and other information.
Sample scene: an AI guide gives a tour of a 3D factory floor in Convai Sim.
Add it all up and your people need two things: help learning the job, and help doing it. Training covers soft skills, practiced in conversation with Avatar Studio, and hard skills, learned on CAD models in Assembly Studio and in 3D simulation in Convai Sim. Assistance covers three places work happens: inside your web and desktop applications, in CAD and 3D workflows, and in physical work with the Vision Assistant.
Training and assistance side by side, with one agent platform underneath.
Picture a plant bringing a new production line online. Shift leads rehearse conversations with their crews in Avatar Studio, while technicians learn each machine from its CAD model in Assembly Studio and walk a replica of the floor in Convai Sim. At the real equipment, the Vision Assistant guides them through each procedure. At the end of the shift, the agent in your web app helps them file the report.
You don't need a separate agent for each product. The agents you build are available across Avatar Studio, Assembly Studio and Convai Sim, and in your own apps through our plugins and SDKs: one core agent, one deployment, with agent management in one place.
Capture, learn, work: the same agents and the same knowledge at every stage.
Assistance matters as much as training. When a question comes up on the job, your people can ask the agent first, and the person everyone asks can get on with their own work.
Every 3D experience has to be rendered somewhere, and you choose where. With AI only, the agent platform and AI models run on your servers, and your own applications and engines render the experience. It suits teams that build in Unity, Unreal Engine or the browser.
Choose the full package and Convai's pixel-streaming infrastructure renders your 3D experiences, such as avatars and Convai Sim scenes, on your own GPU servers and streams them to any browser, so trainees need nothing more than a laptop, tablet or phone.
Two ways to run Convai in your own data center: AI only, or the full package with pixel streaming and rendering engines.
Either way, your applications connect to your own deployment, not Convai's hosted service, so requests and keys stay inside your network. The docs show how to point the Web SDK at your own deployment and, with the full package, how to run pixel streaming on your own servers. We're also making Convai available through AWS Marketplace and Azure Marketplace.
See it in action: AI training simulations and guided tours
Making the case internally? Three demos show the layers working together.
A guided tour of a factory floor
In a factory-floor tour inside a full 3D simulation, the AI guide takes visitors from one location to the next and explains what they're looking at. New hires can walk the line before their first shift on it.
With a partner, we're using Assembly Studio for Mars rover maintenance and repair training: learning a procedure step by step and understanding how the rover is put together. Ask where a component sits, and the guide highlights it on the model.
A Mars rover model in Assembly Studio: asked where the hazard-avoidance cameras are, the guide highlights them on the model.
Lunar training with our partner Buendea
Our partner Buendea, which works with NASA, built a lunar training simulation with Convai. Astronaut trainees practice tasks such as repairing a power station, guided by an AI that sees what they see.
Can Convai run on-premises or in an air-gapped environment?
Yes. You can run the whole platform on your own servers or in your own cloud. Every component ships as a container, nothing leaves your network, and it can run fully air-gapped. Convai is SOC 2 compliant.
Can we use the language models our company already approves?
Yes. Use the models the platform provides, or connect the LLMs your enterprise already approves through an OpenAI-compatible endpoint.
Do Avatar Studio, Assembly Studio and Convai Sim use the same agents?
Yes. The agents you build are available across all three, and in your own apps through our plugins and SDKs: one core agent, one deployment, managed in one place.
Do we need CAD files to use the Vision Assistant?
No. The Vision Assistant also works from the videos, PDFs, manuals and documentation you've already connected. When you do have CAD files, the same AI knows those too.
Do learners need a powerful computer?
No. Avatar Studio and Convai Sim experiences run in a web browser, with nothing to install and no high-end GPU needed on the learner's device. With the full package, the 3D rendering happens on your own GPU servers and streams to the browser.
Start with the questions your experts answer every day
Ask your most experienced people which questions they answer again and again, and which of those answers already sit in a manual, a CAD file or a training video. That list is a good first scope for an agent.
For the whole story in one sitting, watch Convai CEO Purnendu Mukherjee at the Augmented Enterprise Summit 2026.
{"@context":"https://schema.org","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"Can Convai run on-premises or in an air-gapped environment?","acceptedAnswer":{"@type":"Answer","text":"Yes. You can run the whole platform on your own servers or in your own cloud. Every component ships as a container, nothing leaves your network, and it can run fully air-gapped. Convai is SOC 2 compliant."}},{"@type":"Question","name":"Can we use the language models our company already approves?","acceptedAnswer":{"@type":"Answer","text":"Yes. Use the models the platform provides, or connect the LLMs your enterprise already approves through an OpenAI-compatible endpoint."}},{"@type":"Question","name":"Do Avatar Studio, Assembly Studio and Convai Sim use the same agents?","acceptedAnswer":{"@type":"Answer","text":"Yes. The agents you build are available across all three, and in your own apps through our plugins and SDKs: one core agent, one deployment, managed in one place."}},{"@type":"Question","name":"Do we need CAD files to use the Vision Assistant?","acceptedAnswer":{"@type":"Answer","text":"No. The Vision Assistant also works from the videos, PDFs, manuals and documentation you've already connected. When you do have CAD files, the same AI knows those too."}},{"@type":"Question","name":"Do learners need a powerful computer?","acceptedAnswer":{"@type":"Answer","text":"No. Avatar Studio and Convai Sim experiences run in a web browser, with nothing to install and no high-end GPU needed on the learner's device. With the full package, the 3D rendering happens on your own GPU servers and streams to the browser."}}]}
