Beever AI

What Beever AI does

Beever AI Limited is a Toronto-based AI research lab operating as a subsidiary of Hong Kong-headquartered Votee AI Limited. The company focuses on low-resource language LLM training, agentic AI research, and enterprise-grade agent applications, with academic collaborations listed alongside HKUST, Stanford, the Vector Institute, and the University of Waterloo. Its mission centers on culturally-aware AI, illustrated by the HKCanto-Eval Cantonese benchmark and the MAGIC CLI toolchain for low-resource LLM training.

Its flagship product, Beever Atlas, is an open-source knowledge intelligence platform that ingests chat streams from Slack, Discord, Microsoft Teams, and Telegram and converts them into a persistent, queryable memory layer. The platform uses a dual-memory architecture combining Weaviate vector search (handling approximately 80% of queries) with a Neo4j knowledge graph (handling approximately 20% for multi-hop reasoning), orchestrated through a 6-stage AI ingestion pipeline (Sync, Extract, Validate, Store, Cluster, Wiki) driven by Google ADK agents. Gemini 2.0 Flash serves as the primary LLM with Claude fallback via LiteLLM, and the system exposes a REST API and a Model Context Protocol (MCP) server for integration with Claude Desktop, Claude Code, and Cursor. The wider portfolio includes MAGIC CLI for low-resource LLM training, HKCanto-Eval (ACL 2025 CoNLL), and TheoremExplainAgent (ACL 2025 main).

Commercially, Beever AI distributes Beever Atlas under Apache 2.0 as a free open-source product targeting individual developers and teams through GitHub, Docker, and community channels, while a paid Enterprise edition (v0.3.0, planned for Q3 2026; pricing not publicly disclosed) is targeted at regulated organizations requiring SSO/SAML, audit logs, RBAC, compliance reporting, and dedicated support. The company has not disclosed revenue, funding rounds, headcount, or named enterprise customers in available source materials.

Name

Beever AI

Legal name

Beever AI Limited

Website

https://beever.ai

Company type

Private

Operating status

Operating

Short description

Beever AI is a Toronto-based AI research lab (under Votee AI) that builds Beever Atlas, an open-source dual-memory platform converting team chat conversations into a queryable knowledge base, serving engineering and product teams plus regulated enterprises.

Ownership category

akta.pro rank

Product category

Knowledge Management Software

NAICS

Computer Systems Design and Related Services (54151), Computer Systems Design and Related Services (5415)

SIC

Services-Computer Programming Services (7371), Services-Computer Programming, Data Processing, Etc. (7370)

akta.pro primary industry

Question Answering & Retrieval-Augmented Generation (RAG) (HDAAADAC)

akta.pro secondary industries

Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO), Generative AI & LLM Solutions Services (RAG, Agents, Copilots) (BPAEAHAG), Model Development & Training Platforms (AutoML, Notebooks, Feature Stores) (HDAEANAB)

Keywords

• Knowledge management
• AI research lab
• LLM wiki
• Semantic search
• Knowledge graph

Headquarters

HQ city

Toronto

HQ country

Canada

HQ region

North America

Offices2 records

Markets served

GTM type

B2B and B2C

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Operations

Revenue model

1. Open Source (Apache 2.0): Free tier for individual users with full core features including dual memory, wiki generation, multi-platform sync, MCP server. No revenue generated; used for community building and adoption.
2. Enterprise Edition: Paid enterprise tier for regulated organizations requiring high-security deployments. Features include SSO/SAML, audit logs, compliance reporting, role-based access controls, spend controls, dedicated support, SLA, and custom deployment options. Revenue generated through subscription model with direct sales inquiry ([email protected]).
3. Professional Services: Implementation and consulting support for enterprise deployments, likely offered as add-on services for organizations requiring hands-on deployment assistance.

Pricing tiers

Go-to-market motion3 records

Distribution channels4 records

Marketing channels6 records

Core offering

Beever AI develops Beever Atlas, a continuous knowledge intelligence platform that transforms ephemeral chat streams from Slack, Discord, Microsoft Teams, and Telegram into a permanent, structured memory layer using a dual-memory architecture (3-tier semantic store in Weaviate plus knowledge graph in Neo4j). The platform powers an auto-generated LLM Wiki and QA Agents exposed via MCP into Claude Code and Cursor. The lab also produces research tools including MAGIC CLI for low-resource language model training, TheoremExplainAgent for multimodal theorem explanation, and the HKCanto-Eval Cantonese benchmark.

Product overview

Beever AI is a Toronto-based AI research lab under Votee AI, offering a portfolio of products centered on Beever Atlas — a continuous knowledge intelligence platform that transforms team conversations into a living wiki. Beever Atlas uses a dual-memory architecture combining semantic vector search (Weaviate) with knowledge graph (Neo4j) to power automatic wiki generation and QA agents via MCP integration. The portfolio also includes research tools like MAGIC CLI for low-resource LLM training, HKCanto-Eval for Cantonese language benchmarking, and TheoremExplainAgent for multimodal theorem explanations.

Differentiator

Problem solved

Functional benefit

Brands

• Beever Atlas: Open-source LLM Wiki product that transforms team conversations into structured, searchable knowledge using dual-memory architecture (semantic + graph).
• MAGIC CLI
• TheoremExplainAgent
• HKCanto-Eval

Products and services

• Beever Atlas Continuous knowledge intelligence platform that ingests conversations from Slack, Discord, Microsoft Teams, and Telegram, extracts facts, entities and relationships via a 6-stage AI pipeline, and surfaces results through an LLM Wiki and QA Agents with MCP integration for Claude Code and Cursor. Offered as an Apache 2.0 open source edition for individuals and developers and a paid Enterprise edition for regulated organizations.
• MAGIC CLI Advanced agent team (command-line tool) that facilitates training of low-resource language models and industrial LLMs, offering configurable infrastructures and comprehensive data source integration.
• HKCanto-Eval Pioneering benchmark for assessing LLMs on Cantonese language understanding and Hong Kong's cultural nuances, providing a comprehensive framework for culture-aware language models.
• TheoremExplainAgent Multimodal approach for enhancing how LLMs comprehend and explain complex theorems, producing text-to-long-video outputs for STEM education.

Named customers1 record

Segments5 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration7 records

AI capability5 records

Feature8 records

Partnerships

13 partnerships are on record, tiered core, minor and supporting.

• Votee AIcoreStrategic or Co-development PartnerHong Kong-based enterprise AI company that founded Beever AI as its Toronto-based research lab. Votee AI operates as the parent company, with Beever AI specializing in low-resource languages LLM training, agentic AI research, and enterprise-grade agent applications. Beever Atlas is co-developed and open-sourced by both entities.
• Hong Kong University of Science and Technology (HKUST)minorStrategic or Co-development PartnerAcademic research collaboration. HKUST is listed among university collaborators including Stanford, Vector Institute, and University of Waterloo. Logo displayed in collaboration section.
• Vector InstituteminorStrategic or Co-development PartnerAI research institute collaboration. Located in Toronto alongside Beever AI's research lab. Logo displayed in collaboration section with other academic institutions.
• Stanford UniversityminorStrategic or Co-development PartnerAcademic research collaboration. Stanford logo displayed alongside HKUST, Vector Institute, and University of Waterloo in collaboration section.
• University of WaterloominorStrategic or Co-development PartnerAcademic research collaboration. Waterloo logo displayed alongside other university partners. Contributes to research pipeline and talent acquisition.
• WeaviatecoreTechnology or IntegrationVector database powering semantic memory. Used for vector embeddings and semantic search, handling ~80% of queries. Part of core dual-memory architecture alongside Neo4j graph store.
• Neo4jcoreTechnology or IntegrationGraph database powering knowledge graph memory. Used for entity extraction, relationship mapping, and multi-hop reasoning. Handles ~20% of queries requiring relational context.
• Google (Google ADK)coreTechnology or IntegrationAgent Development Kit used to orchestrate multi-stage ingestion and query pipeline. Google ADK agents handle extraction, validation, routing, and generation tasks.
• Jina AIcoreTechnology or IntegrationJina v4 (2048-dimension) multimodal embeddings used for semantic search. Powers vector representations for facts, topics, and multimodal content.
• Anthropic (Claude via LiteLLM)coreTechnology or IntegrationClaude models used as fallback via LiteLLM when Gemini is unavailable. Part of multi-model support for LLM synthesis, fact extraction, and response generation.
• Google (Gemini)coreTechnology or IntegrationGemini 2.0 Flash used as primary LLM for fact extraction, routing, and response generation. Part of dual-model strategy with Claude fallback.
• VercelsupportingTechnology or IntegrationVercel Chat SDK (TypeScript) used for real-time message capture from chat platforms. Powers the bot framework for platform adapters.
• TavilysupportingTechnology or IntegrationExternal web search used for best practices and documentation retrieval. Part of query pipeline for comprehensive answers.

Scale indicators4 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Khoj: Open-source personal AI copilot that ingests documents, notes, and conversations to enable semantic search and chat. Comparable self-hosted, open-source positioning to Beever Atlas with overlapping memory/retrieval architecture.
• Mem0: Memory layer for AI applications and agents, focusing on persistent, queryable memory across conversations. Direct functional overlap with Beever Atlas's RAG/memory capability for teams and developers.
• Haystack (deepset): Open-source NLP framework for building production RAG pipelines, question answering, and semantic search systems. Competes at the developer framework layer with overlapping pipeline architecture (ingest, retrieve, generate).
• LlamaIndex: Framework for building RAG and agentic applications over custom data sources. Overlaps with Beever Atlas on the underlying ingestion/retrieval pipeline architecture and developer audience.
• Glean: Enterprise AI search and knowledge management platform that connects to workplace apps for unified knowledge retrieval. Directly competes for the same regulated-enterprise buyer that Beever targets with its Enterprise edition.

Others

• Obsidian: Personal knowledge management tool with linked notes and graph view; often used as the manual alternative Beever Atlas positions against for wiki curation. Adjacent knowledge management category with strong community.

Emerging players

• Hebbia: AI knowledge worker for enterprises that ingests documents and conversations to answer complex questions. Comparable enterprise knowledge retrieval positioning with citations and grounded answers.
• Contextual AI: Enterprise RAG platform purpose-built for grounded AI applications with similar focus on accuracy, citations, and enterprise security. Comparable positioning for regulated enterprise RAG deployments.

Broad incumbents

• Microsoft Copilot: Microsoft's enterprise AI assistant integrated across Microsoft 365 including Teams. Competes broadly with Beever's Teams integration and enterprise knowledge management positioning.
• Notion AI: AI features embedded in the Notion workspace product, including wiki generation, Q&A, and summarization. Larger incumbent overlapping Beever's automatic wiki generation and team knowledge use cases.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat6 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/Beever-AI/beever-atlas
• Xx.com/Beever_AI
• LinkedInlinkedin.com/company/beever-ai

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Beever AI