Votee AI's Beever Atlas Turns Chat Chaos into Sovereign Knowledge

By Sam Lidman

TORONTO and HONG KONG – May 08, 2026 – A Hong Kong AI company and its Toronto-based research lab have launched an ambitious open-source platform designed to solve a problem plaguing modern workplaces: the vast amount of knowledge lost in the constant stream of team chat. Today, Votee AI and its lab Beever AI released Beever Atlas, a tool that transforms conversations from Slack, Microsoft Teams, Discord, and other platforms into a structured, searchable, and AI-ready knowledge base.

The launch is positioned as a direct answer to a challenge issued by Andrej Karpathy, a founding member of OpenAI, who called for an “incredible new product” to manage knowledge for large language models (LLMs). By automatically cataloging the unstructured data of daily work conversations, Beever Atlas aims to create a persistent and compounding organizational asset from what is often an ephemeral resource.

Answering the Call for a Smarter AI Memory

The AI industry has been grappling with how to provide LLMs with reliable, evolving, and context-rich memory. In a viral social media post that drew tens of millions of views, Karpathy argued that LLMs need more than just raw data feeds or simple vector similarity searches; they need structured knowledge. His proposed prototype involved a largely manual, single-user process of feeding curated files to an LLM to build a personal wiki.

Beever Atlas takes this concept and applies it to the enterprise, starting not with files, but with the primary source of collaborative work: team chat. “Every growing organization faces the same silent liability: conversational knowledge loss,” said Pak-Sun Ting, Co-Founder and CEO of Votee AI. “Beever Atlas turns this perishable resource into a compounding organizational asset.”

The platform fundamentally differs from Karpathy's initial local approach. It offers chat-native ingestion across five major platforms without manual uploads, a zero-install web UI, and a multi-user architecture built for teams. It also goes beyond text, unifying images, voice, video, and PDFs into a single memory layer. This approach aims to capture the complete context of organizational knowledge, which is often fragmented across different media types and conversations.

The Technical Bet: Structure Over Similarity

At its core, Beever Atlas makes a significant technical bet: that for an AI to be truly useful to an organization, a deep understanding of relationships is more valuable than simply finding similar words. While many AI systems rely on Retrieval Augmented Generation (RAG) and vector similarity search—essentially finding text that looks like the user’s query—Beever Atlas focuses on building a structured knowledge graph.

Using Neo4j, the platform maps out typed entity relationships between people, projects, technologies, and key decisions discussed in chat. “The key technical decision was to treat agent memory as a knowledge engineering problem, not a retrieval problem,” explained Jacky Chan, Co-Founder and CTO of Votee AI. “Structure beats similarity—a typed graph of who works on what is more useful to an AI than vector search over a Slack archive.”

This structured graph then serves as the memory layer for any AI assistant. Beever Atlas ships with a native MCP (Memory, Cognition, Perception) server, allowing coding assistants like Cursor, AWS Kiro, and Qwen Code to query the team's collective knowledge directly. This transforms the AI from a general-purpose tool into a specialized team member with full context on the organization's history and activities. Further integrations for the open-source edition with emerging tools like OpenClaw and Hermes Agent are planned for Q2 2026, solidifying its role as a foundational knowledge backend.

A New Frontier in Sovereign AI

The launch of Beever Atlas also highlights Votee AI's focus on a critical, emerging field: Sovereign AI. This concept refers to an organization's or nation's ability to maintain control over its AI technology stack, data, and models, ensuring compliance with data protection regulations and preventing reliance on external entities. For regulated industries like finance and government, this is not just a preference but a requirement.

Beever Atlas is built from the ground up for data sovereignty. The entire system runs as a Docker stack within the customer's own on-premise or cloud environment, ensuring data never leaves their perimeter. It features a “Bring Your Own LLM” model, allowing organizations to use their own locally run models via Ollama or connect to over 100 supported cloud-based LLMs. With zero telemetry and AES-256-GCM encryption at rest, the platform is designed for maximum security.

This project is part of a broader strategy for Votee AI, which has established itself as a key player in Asia's sovereign AI landscape. The company previously developed the first fully pre-trained open-source Cantonese LLM and validated its platform in the Hong Kong Monetary Authority's highly regulated fintech sandbox. “Hong Kong has always been known for property and finance,” Ting stated. “Beever Atlas is proof that world-class AI infrastructure can emerge from an HK-headquartered company and be shared openly with the world.”

Built for the Enterprise: Security and Compliance by Design

While an Apache 2.0 open-source edition is available for individuals and developers, the Enterprise Edition of Beever Atlas includes a suite of features specifically designed for large, regulated organizations.

Perhaps the most critical of these is “Permission Mirroring.” A common fear with enterprise AI is that it might inadvertently leak sensitive information by drawing on data from private channels. Beever Atlas solves this by exactly mirroring the permissions from platforms like Slack and Microsoft Teams. If a user doesn't have access to a private channel, the AI cannot use information from that channel to answer their questions, with permission changes propagating in under a minute.

Other enterprise-grade features address the full lifecycle of data governance and security. Identity management is handled through SSO and SCIM via Okta or Google Workspace, ensuring access is tied to an employee's status. Immutable audit logs provide a tamper-evident record of all queries and actions for compliance, while configurable data retention policies automatically purge old data. The platform also includes built-in prompt-injection defense to prevent malicious manipulation and context federation to connect with other enterprise systems like Salesforce and Jira, creating a holistic view of organizational knowledge.

Beever Atlas is available immediately on GitHub for self-hosting, with a managed cloud version planned for the second half of 2026. By tackling the chaotic, unstructured world of team chat, Votee AI is not just creating a wiki; it's building a comprehensive, secure, and intelligent memory for the modern enterprise.