Hong Kong-based Votee AI, in partnership with its Toronto research lab Beever AI, has announced the open-source release of Beever Atlas. Responding to the industry's call for structured "LLM Knowledge Bases"—notably championed by AI pioneer Andrej Karpathy—Beever Atlas automatically converts unstructured conversations from Telegram, Discord, Mattermost, Microsoft Teams, and Slack into a structured Neo4j knowledge graph. This "memory layer" provides AI assistants with a typed, searchable, and evolving understanding of organizational intelligence, moving beyond the limitations of simple vector search.
Chat-Native Knowledge: Automatically ingests and structures team conversations across five major chat platforms.
Structured Over Similarity: Uses a Neo4j knowledge graph to map relationships between people, projects, and decisions, rather than relying solely on vector similarity.
Dual Editions: Available as an Apache 2.0 Open Source Edition for individuals and a high-security Enterprise Edition for regulated organizations.
Agent-Ready Memory: Includes a native Model Context Protocol (MCP) server, allowing tools like Cursor, AWS Kiro, and Qwen Code to query team knowledge directly.
Sovereign & Secure: Runs 100% on-premise as a Docker stack; features "Bring Your Own LLM" via LiteLLM and Ollama.
Multimodal Intelligence: Unifies text, voice, video, images, and PDFs into a single searchable memory layer.
Most organizational knowledge is created in chat but remains inaccessible to traditional AI tools. Beever Atlas is designed to turn this perishable resource into a compounding asset. Unlike manual, file-based prototypes, Beever Atlas provides a zero-install web UI that continuously updates a "living wiki" of team intelligence.
"Hong Kong has always been known for property and finance," said Pak-Sun Ting, Co-Founder and CEO of Votee AI. "Beever Atlas is proof that world-class AI infrastructure can emerge from an HK-headquartered company and be shared openly with the world."
The Enterprise Edition is purpose-built for banks and government agencies with stringent security requirements. It addresses the "Don't Leak Secrets" challenge through Permission Mirroring, ensuring the AI only answers based on channels the user is authorized to see.
| Feature | Enterprise Capability |
| Permission Mirroring | Reflects Slack/Teams permissions; updates propagate in <60 seconds. |
| Identity Management | SSO + SCIM via Okta or Google Workspace with hard data isolation. |
| Audit & Compliance | Immutable audit logs and configurable data retention policies. |
| Trust & Safety | Built-in prompt-injection defense and hallucination checks. |
| BYOC Deployment | "Bring Your Own Cloud" (AWS/Azure) to keep data within the customer perimeter. |
The Technical Bet: Structure Beats Similarity
The development of Beever Atlas was led by Jacky Chan, creator of the first fully pre-trained open-source Cantonese LLM. The team’s core thesis is that a typed graph of entity relationships is fundamentally more useful for AI reasoning than raw text chunks.
"The key technical decision was to treat agent memory as a knowledge engineering problem, not a retrieval problem," said Chan. Looking ahead to Q2 2026, the project will ship dedicated updates for OpenClaw and Hermes Agent, making Beever Atlas one of the first MCP-native backends optimized for these autonomous workflows.
About Votee AI
Votee AI (Votee Limited) is a Hong Kong-headquartered enterprise AI company with a research lab, Beever AI, based in Toronto. Specialized in sovereign AI infrastructure, Votee AI has delivered breakthrough open-source LLMs and validated its security through the Hong Kong Monetary Authority’s FSS 3.1 Pilot programme.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.