Context Engine
Deep understanding of your project structure, codebase, and working context for intelligent assistance.
In development · Early access opening soon
Autonomous AI agents and natural voice AI, built for real work.
Dantrox is an AI research startup from Vietnam building two products: Dx Agentic, agents designed to run complex workflows end to end, and Dx Voice, speech with human warmth in Vietnamese and other languages.
Dx Agentic
- Read 128 new support tickets0:12
- Grouped them into 3 root causes0:41
- Drafted fixes and opened 3 pull requests1:38
- Briefed the team by voice, in Vietnamese2:14
Dx Voice · Linh
“Đã xử lý xong ba sự cố. Các bản sửa đang chờ anh chị duyệt.”
A research-driven AI startup from Vietnam. We are building autonomous agents and voice systems that close the gap between what large language models promise and what production demands.
Research-first
We don’t just wrap existing models. We research and fine-tune to solve real limitations in current AI systems.
Built for production
Designed from the ground up for enterprise reliability, security, and scalability.
Local excellence, global standard
Deep understanding of Vietnamese language and business context, backed by rigorous AI research and engineering.
Affordable intelligence
We aim to make capable AI affordable and quick to adopt for growing companies.
In development
Dx Agentic
Scalable autonomous AI agents built for complex workflows.
Dx Agentic is an agent platform we are building to understand, plan, and execute complex business processes with minimal human intervention. It is grounded in our research into LLM limitations, with the goal of more reliable reasoning, better tool integration, and consistent performance in production.
- Autonomous multi-step workflow execution
- Advanced reasoning and error recovery
- Seamless integration with existing tools and systems
- Designed for enterprise reliability and security
# general— Swarm active on v2.4.1 release pipeline
4 active nodes
Illustrative preview of the Dx Agentic interface. Agents, data and messages are fictional.
In development
Dx Voice
Natural voice AI with emotional expressiveness.
Dx Voice is a text-to-speech and voice cloning platform in development. Our current model is built on a fine-tuned CosyVoice 3.0 model and designed for natural, emotionally expressive speech and zero-shot voice cloning.
- High-fidelity voice cloning with just seconds of audio
- A strong focus on Vietnamese and multilingual speech
- Expressive, consistent, and human-like speech
- Built for virtual assistants, customer service, and media
Open Source License
The Dx Voice model is fine-tuned from CosyVoice 3.0, developed by FunAudioLLM and licensed under the Apache License 2.0. Dantrox is not affiliated with or endorsed by FunAudioLLM.
EnChinaVietJapanMore languages
0:12
Warm assistant voice for calm product narration.
“Good morning. Your calendar has three meetings today, starting with the design review at 10 AM.”
0:12
Smooth Mandarin voice for commerce and service flows.
“早上好。您的订单已经确认，预计今天下午送达。”
0:11
Natural Vietnamese voice for assistants and support UX.
“Xin chào! Bạn có một tin nhắn mới từ nhóm phát triển sản phẩm.”
0:14
Clear, composed voice for polished multilingual agents.
“おはようございます。今日のスケジュールを確認しましょう。”
Sample outputs from our current Dx Voice model.
What we are building. Every capability is grounded in our research and designed for real-world reliability.
Agentic Execution
Autonomous task completion with multi-step reasoning, tool usage, and self-correction capabilities.
Voice Synthesis
Neural TTS with natural prosody, emotional expression, and real-time streaming in multiple languages.
Privacy by Design
Designed so sensitive data can stay under your control, with encryption in transit and at rest.
API First
Planned REST APIs and SDKs for integration into existing workflows and third-party applications.
Continuous Learning
Models designed to improve with usage while keeping strict privacy boundaries.
Frequently asked questions. How we build reliable agents and natural voice AI.
Dx Agentic is being designed around multi-step reasoning, self-correction loops, and fallback paths, using specialized fine-tuned models rather than general-purpose APIs. Our aim is lower latency and fewer hallucinations in production.
Dx Voice is built on a CosyVoice 3.0 model that we fine-tune for expressive speech. Our goal is speech that conveys emotions such as excitement, empathy, and calm, and zero-shot voice cloning from a few seconds of reference audio. The samples on this page come from our current model.
Security is a core design requirement. We are designing for on-premise deployment options, role-based access control (RBAC), and encryption of data in transit and at rest, and we do not intend to use customers' proprietary data or conversation logs to train public models. Specific commitments will be set out in our terms at launch.
Both products are in private development. We are opening early access to a small number of teams. Email ct@dantrox.com to join the list, and we will share timelines and onboarding details directly.
Get early access to Dantrox.
Dx Agentic and Dx Voice are in private development. Tell us about your use case and we’ll reach out as early access opens.