Convai launches on-premise AI platform for training and workplace assistance

One agent for training and daily work

Convai says workplace questions often return to the same internal experts, creating interruptions for the worker seeking help and the specialist providing it. Its platform is designed to address both formal training and the less structured learning that happens during daily work, such as at a machine, inside an application or during a difficult conversation.

The company says the answers employees need often already exist in CAD files, manuals, procedures and training videos. Convai’s platform processes documents, videos and other files into multimodal knowledge that agents can use in conversations. The same agent and knowledge can support a person learning a procedure in a 3D simulation and later carrying it out on real equipment.

Convai positions the platform for environments such as aerospace, defense and manufacturing, where company knowledge may need to remain inside an organization’s network. Customers can build agents on their own servers or in their own cloud and deploy them in applications, 3D simulations and devices used by employees.

Applications span web, 3D and physical work

The platform includes a Web SDK for adding agents to web applications and portals. These agents can interview users, fill in long forms, highlight parts of an interface and guide people through tasks. Convai also offers SDKs for Unity, Unreal Engine, NVIDIA Omniverse and three.js, allowing conversational agents to be added to existing 3D simulations.

Avatar Studio supports conversational avatars for trade-show and shop-floor kiosks, as well as practice sessions for sales conversations, HR situations and leadership coaching. Assembly Studio uses CAD files to help people ask about parts and assemblies by voice, while generating step-by-step procedures for experts to review.

Convai Sim combines Gaussian-splat captures and digital twins with CAD knowledge, explorable 3D environments and conversational avatars. Vision Assistant runs on a phone or tablet, or in an XR application built with Convai’s plugins, allowing an AI system to see what a technician sees and guide each step using manuals and documents.

On-premise infrastructure and access controls

Convai describes the platform as a stack that begins with a customer’s AI data center and includes agent infrastructure, AI models, SDKs and products. The foundation provides agent creation, live voice and text conversations, scaling across the customer’s GPU servers, observability and authentication.

Access is controlled through directory groups, MCP servers and credentials approved by the customer. Convai says an API key stays on the customer’s server, while browsers use tokens issued by the customer. Agents can connect to internal tools and data through MCP servers with OAuth authorization.

The platform also includes agent management, a Knowledge Bank, long-term memory, Mindview and team workspaces. Mindview shows the full prompt behind a response so teams can trace and fix unexpected answers. A full package adds pixel-streaming infrastructure, including matchmaking servers, signalling and proxy services, asset management and stream scaling, allowing 3D experiences such as avatars and Convai Sim to stream from customer servers to users’ browsers.

For users, the announcement describes a single deployment that can extend from training environments to web applications, simulations and physical-work assistance. For organizations, Convai’s model emphasizes retaining control of data, credentials, rendering and deployment while using the same agent across multiple work settings.