Maum AI 마음AI
KOSDAQ-listed Korean AI firm building robot brains and edge AI models
Overview
Maum AI (마음AI), formerly MINDs Lab, is a KOSDAQ-listed Korean company that now focuses on Physical AI: robot and machine control built on its MAAL, SUDA, BODA and WoRV models and its MAIED edge module. Its maum.ai platform still sells speech, chatbot, voicebot and AI-human services on a published won price list, with robot projects quoted individually.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- All platform services for a month
- Orchestra allowance: TTS 10,000, STT 10,000, GPT 20,000, STF 1,000 characters
- TTS, STT and scenario chatbot ₩1 per Korean character
- maumGPT-13B ₩100 per 1,000 characters
- STF with TTS ₩10 per character
- Voicebot ₩1,000 per successful call
- Visible voicebot ₩800 per successful call
- M3 AI Human ₩990,000 per device a month
- TTS ₩2M (emotional ₩5M)
- STT ₩20M with 300 hours of data
- STF ₩30M-₩35M; voicebot ₩25M per scenario
- LLM inference model and maum.ai platform operation
- Usage charged at inference rates
- AI consulting and fine-tuning
- Maum Academy training
Use cases
What you can produce with Maum AI 마음AI
- Robots and machines given on-device perception and control through the MAIED edge module
- Korean speech synthesis and transcription billed at ₩1 per character
- Phone voicebots charged per successful call
- Quadruped security and inspection robots managed from one control system across manufacturers
- Custom TTS, STT and chatbot engines trained on a company's own data
- Open MAAL and CANVAS models downloaded from Hugging Face for research
ASEAN Perspective
Maum AI 마음AI in Southeast Asia
Maum AI has an established office presence in the Philippines — a natural beachhead given that country's large BPO and contact-centre industry — and lists Southeast Asia as a target market for AICC and AI English education. However, its customer references and publicly documented deployments remain concentrated in South Korea (banking, telecoms, public sector). Korean-language optimisation is a core competitive strength, which reduces out-of-the-box utility for non-Korean ASEAN deployments; organisations in the region would need to evaluate multilingual or English-language pipeline capabilities separately. The company's Physical AI Data Factory and WoRV robotics work are aligned with ASEAN industrial automation growth, but commercial traction in the region has not yet been publicly confirmed as of mid-2026.
Maum AI pairs genuine research, including the award-winning CANVAS navigation system and open MAAL models on Hugging Face, with the transparency of a listed company. Its Physical AI work has real partners, from Doosan Bobcat to a national autonomous-ship project, and its maum.ai speech and chatbot services carry published per-character prices.
The financials are the caution: losses continue and first-half 2026 revenue fell sharply. For ASEAN buyers its Thai security-robot and mine-detection trials are the relevant track record; treat it as a specialist for pilots, not a mature global vendor.
What we found
Maum AI (마음AI), formerly MINDs Lab, is a Korean AI company based in Pangyo, Seongnam, operating since January 2014 and listed on KOSDAQ under ticker 377480 since November 2021. Its homepage now calls it a Physical AI company building robot brains, and it claims to be the only Korean company to have commercialised Physical AI, a claim that cannot be tested from outside. Its four in-house model families are MAAL, an on-device agent language model; SUDA, a combined speech recognition, LLM and speech synthesis model; BODA, a small vision-language model; and WoRV, a vision-based robot control model. They run on MAIED, its edge AI module.
The research is real. CANVAS, its commonsense robot-navigation system, won an Outstanding Paper award at a NeurIPS 2024 workshop and was accepted at ICRA 2025, and MAAL-8B and MAAL-11B-Vision are published on Hugging Face. In June 2026 Maum AI signed an MOU with Doosan Bobcat on autonomous compact construction equipment, and in September it was named lead of a ₩26.2 billion national project on emergency response for autonomous ships, running to 2032.
Its traceable Southeast Asian work is in Thailand: a security-robot project with the Hatari group since July 2025, with a Bangkok demonstration planned for August 2026, and a quadruped mine-detection system built for Thai field trials.
The filings are less flattering. Revenue was ₩9.8 billion in 2025, up from ₩7.8 billion, with an operating loss of ₩6.4 billion, and the first half of 2026 brought only ₩2.7 billion of revenue against a ₩5.1 billion operating loss. The older maum.ai platform still publishes a won price list for speech, chatbot and AI-human services. Robots and edge deployments are quoted per project. With about 130 staff, it is a specialist rather than a platform giant, and most documentation and support are in Korean, so non-Korean buyers should plan for a partner-led pilot.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Notable facts
- Maum AI's VoiceFilter repo — an unofficial PyTorch implementation of Google's audio separation research — has accumulated over 1,200 GitHub stars, making it one of South Korea's most-starred AI open-source projects.
- CEO Yoo Taejoon holds a CPA license from Samil PwC, making him one of the few AI company founders whose career began in accounting rather than engineering.
- Maum AI's CANVAS robot-navigation system won an Outstanding Paper Award at the NeurIPS 2024 Workshop on Open-World Agents, placing it among a small group of Korean AI labs with top-tier international workshop recognition.
- The company's WoRV division (World Models for Robotics and Vehicle Control) operates a fleet of hundreds of robots in real industrial sites — agricultural, logistics, defense, and construction — as training data generators for its foundation models.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Maum AI 마음AI's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Maum AI 마음AI unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Maum AI 마음AI directly →
Spotted something out of date? Suggest an update →
Alternatives to Maum AI 마음AI
More in Business & Finance
