Enterprise artificial intelligence company Articul8 has launched Bivia, a multimodal decision model designed to help AI agents evaluate information, verify evidence and make decisions across business workflows without depending on large language models for every task.
The model is available through an application programming interface (API) and can operate independently or alongside Articul8's existing enterprise AI platform. It is intended to support organisations deploying AI agents for tasks that require repeated assessments of information from different sources.
Bivia processes text, images, documents, tables, charts, audio and video. Unlike generative AI models primarily used to produce content, the system focuses on evaluating available information and determining the next action within a workflow.
The model can assist AI agents in selecting appropriate tools, checking whether information is consistent, identifying supporting evidence and determining whether an assigned task has been completed.
Ananthanarayanan S, Managing Director, India at Articul8, said, "Enterprise AI is moving beyond generating information to taking action, and that shift makes decision-making just as important as generation."
According to the company, Bivia can be customised using an organisation's internal data, industry knowledge, operational policies, standard operating procedures and previous decisions.
It supports five categories of decision-making: Choice, Boolean, Score, Set and Span. These functions allow AI agents to select between options, verify statements, assign scores, identify relevant categories and locate evidence within source material.
Articul8 has introduced three operating modes for the model. Fast is intended for routine decisions, Fast+think provides additional processing for more complex assessments, while Deep is designed for tasks requiring further verification.
Bivia can also withhold a decision when the available information is insufficient, rather than producing a response without adequate supporting evidence.
The company said the model supports multilingual applications, including Hindi, Gujarati, Tamil and Sanskrit. Its multimodal capabilities also allow it to identify relevant sections of documents, regions within images and timestamps in video recordings.
In internal evaluations, Articul8 reported near-perfect accuracy on selected energy and hardware design tasks after domain-specific training. The company also said Bivia completed approximately 70% of workflows in one evaluation, compared with around 30% for the alternative system tested.
These performance figures are based on company-reported assessments and have not been independently verified.
Articul8 demonstrated the model's processing capabilities using a 52-minute, 4K documentary filmed aboard the International Space Station, analysing its video frames in a single request.
The launch expands Articul8's enterprise AI offerings as organisations explore agent-based systems that can perform tasks involving multiple decisions. Bivia is designed to handle evaluation and verification within these workflows, while generative models continue to support functions such as planning, writing and explanations.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.