XGEN —
the Agentic AI Platform for enterprise AX

XGEN is an on-premise Enterprise AI platform — an agent development environment for designing, deploying, operating, and governing Agentic AI services on the LLMs and infrastructure you choose. It is not a finished service delivered to you; you build and run the agents your organization's work needs on top of XGEN. It was designed and developed by the Plateer AI Labs AI research team.

You are not tied to a particular model. Choose among open-weight models such as Qwen and our own sLLM (Polar) by purpose, cost, and accuracy, and serve them on the GPUs and infrastructure you already own. The management center configures the LLM, vector DB, and the rest of the system in detail.

Past simple text extraction (OCR), a VLM (vision-language model) reads the context of images, tables, charts, and graphs. OCR is one capability within the VLM; where needed, an open-weight VLM is served directly to turn charts and graphs into text, vectorize them, and use them in answers. This is already applied in commerce and finance settings.

Yes. Even without tidy data, accumulating a few months of it and constructing an ontology links the different expressions of the same entity (for example US, USA, United States) into one. The richer the data the better the quality gets, and domain terminology mapping and review-rule design happen alongside. We have introduced knowledge graphs in commerce and financial projects.

Quality is scored on RAGAS from preprocessing through answer verification and evaluated across several dimensions. Preprocessing strategy is also selected automatically by document type — an adaptive pipeline where the LLM judges the document type and adjusts whether OCR or VLM applies, along with chunking and overlap parameters.

You don't need to build new screens. LLM and RAG integration is composed as nodes and provided as snippet code, so adding a single button to an existing system to call it is entirely workable. XGEN is plugin-structured throughout — around 80 built-in tools combined with quickly-built custom tools keep integration and customization effort to a minimum.

A first filter applies at creation through risk assessment and administrator approval, and if something goes wrong in operation an admin kill switch stops the service immediately. Where accuracy is decisive we apply human-in-the-loop, designing the AI to raise throughput through pre- and post-processing rather than replacing the work entirely, with a person confirming the result before the next step.

It is built on-premise so data never leaves, and it supports network-separated and air-gapped environments. Role- and attribute-based access control (RBAC/ABAC) plus a zero-training guarantee technically prevent customer data from being used to train external models.

Yes. Agentflow's drag-based visual canvas lets you design a workflow without code, validate it in live chat, and deploy it as an API or workflow.

The open source that makes up XGEN is MIT, Apache 2.0, and BSD family — no GPL-family components with source-disclosure obligations, so commercial and internal distribution carries no obligation to publish source. We check and provide the obligations attached to each license type (copyright notices, NOTICE files), and can prepare a component-by-component license inventory on request.