maum.ai
Maum.AI Inc. (formerly MINDs Lab) · Other
maum.ai is Maum.AI's enterprise AI platform for Korean language models, voice assistants, chatbots, and AI humans, deployed in the cloud, on-premise, or on-device.
About maum.ai
What Is maum.ai
maum.ai is the AI platform built by Maum.AI Inc., a Seoul company formerly known as MINDs Lab. It sells AI as a service: you pick the pieces you need, such as a Korean language model, a voice conversation system, a chatbot, or a digital receptionist, and the company assembles them into a working service for your business.
The platform is organized around foundation modules rather than one big product. MAAL handles language, SUDA handles spoken conversation, WoRV covers movement for robots and vehicles, and BODA covers vision. Above those sit finished applications: AI human receptionists, an AI contact center, a meeting transcription service, and a cloud API.
The main draw is deployment control. Banks, public agencies, and defense units often can't send customer audio to a US cloud, so Maum.AI ships the same models in a private data center or directly on a device. The tradeoff is that nothing here is self-serve. Pricing comes from a quote, setup involves the vendor's team, and the documentation and demos lean heavily on Korean.
Getting Started
- Describe the job on the maum.ai site or through the sales contact form. Typical entries are a call center queue, a lobby kiosk, or an offline voice interface for hardware.
- Choose the modules and the deployment target. Cloud runs fastest to launch, on-premise suits regulated data, and on-device fits products with no reliable network.
- Send sample audio, documents, or dialogue scripts. The team builds the flow in M Studio or AI Human Builder and tunes the speech and language models on your data.
- Connect through the cloud API or an on-site install, then run a pilot on one queue, one floor, or one device before rolling out wider.
Product Information
A quick look at maum.ai's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- Contact center managers in Korea
- Public agencies and regulated teams
- Hardware and robotics product teams
Tasks
- Korean customer service automation
- Meeting notes
- Lobby and front desk reception
- On-premise LLM work
- Voice command for devices
Scenarios
- A hotel adds a multilingual concierge screen without hiring night-shift staff for six languages.
- A bank keeps chatbot logs and call recordings in its own data center while still using a modern language model.
- A logistics robot takes spoken instructions on a warehouse floor where Wi-Fi drops between aisles.
Key features
Korean Language Models Built In-House
MAAL is Maum.AI's own Korean LLM line, and the company positions it for text where imported models often stumble on honorifics, spacing, and domain jargon. Published materials list a 70B on-premise model and a smaller on-device model called Mona. For a Korean insurer or hospital, that local tuning matters more than a leaderboard score.
SUDA Voice Conversation Stack
SUDA is the name of the voice conversation system, and it bundles speech recognition, a language model, and speech synthesis into one loop. Because the whole chain runs on the device, a reply doesn't wait for a round trip to a server. Maum.AI demonstrated it as a tactical voice command system for defense units operating with limited or no network.
AI Humans for Front Desk Work
The AI Human line covers a builder for creating an avatar, video production, a chatbot version, and a physical display device. The company's hotel concierge example answers questions about reservations, facilities, packages, and weather in Korean, English, Japanese, Chinese, Spanish, and French. It suits repeated questions, not open-ended negotiation.
AI Contact Center
The FAST AI contact center product targets the repetitive share of inbound calls. New signups get the first month free, and billing follows actual consultations rather than seats. That structure lowers the cost of a trial, but your bill still moves with call volume.
Cloud API for Speech and Language
Developers can call speech to text, text to speech, chatbot, and machine reading comprehension endpoints instead of building models. The company describes its MRC engine as tuned specifically for Korean, which is the practical reason to pick it over a general API. Rate limits and quotas come with the contract.
Edge and Robot Deployment
Maum.AI was the first company to tune the NPU on Qualcomm's QCS6490 chip, according to its own announcement, and it runs the Qualcomm IoT Accelerator program as Korean partner. WoRV extends the same stack to movement for robots and vehicles, while BODA adds vision. The point is a single vendor for sensing, speech, and motion on constrained hardware.
Pros and cons
Pros
- Korean language quality is the core product, not an afterthought, so honorifics and local business terms get real attention.
- Three deployment paths cover cloud speed, on-premise compliance, and offline devices under one contract.
- Finished applications sit on the same stack, which shortens the path from a model license to a working reception desk or call queue.
- Edge work on Qualcomm hardware gives hardware teams a voice stack that fits a power and heat budget.
Cons
- No public price list, so comparing against a self-serve API takes a sales conversation before you know the number.
- Setup depends on the vendor's engineers, which slows down teams that expect to launch from a dashboard in a week.
- Documentation, demos, and most case studies are in Korean, and international buyers will need local staff to evaluate them.
- The free offer is limited to a first month on the contact center, so there's no permanent free tier to prototype on.
Frequently asked questions
It builds AI services for businesses, mainly in Korean. Common uses are automated call handling, lobby reception with an on-screen AI human, meeting transcription, and voice control for devices and robots.
Related content
Explore related tools, skills, and articles for maum.ai.
maum.ai Alternatives
Odyssey 3
Odyssey · OtherOdyssey-3 is Odyssey's foundation world model. It simulates physics in real time, turns a prompt into an interactive environment, and adapts to robots, cars and games.
JarvisCore
Prescott Data · Coding · Productivity · OtherJarvisCore is an open source Python runtime where AI agents work as equal peers on an agent mesh, claim tasks from a shared ledger, and reach external services through a zero-trust broker.
Clef
Cloudflare · Business · OtherClef is Cloudflare's open-source decision model for Workers AI. It returns calibrated probabilities for typed questions, not generated text, so agents can classify, route, and act fast.