Deepractice: Building the 'JVM for AI Agents' — an Open-Source Platform Giving Every Industry Its Own AI Employee

Forum topic · ✨步子哥 · 2026-01-09

Summary

Deepractice is a Hong Kong-based startup founded in 2025 that aims to become a universal platform for AI agents — a 'virtual machine' that lets anyone in customer service, sales, education, or software development create a personal AI employee simply by chatting. Its core open-source stack includes PromptX (a context platform with 3,359+ GitHub stars), DPML (a declarative configuration language), and AgentX (a next-generation agent runtime). The article contrasts AI agents with chatbots, argues agents that autonomously plan and execute tasks will transform every industry, and cites explosive market forecasts: the global AI agent market is projected to grow from roughly $760-780 million in 2025 to $4.8-5.3 billion by 2030 at a 43-46% CAGR, with optimistic estimates reaching $18.3 billion by 2033. Deepractice Cloud uses a pay-as-you-go model with free tier, Pro subscriptions, enterprise self-hosting, and a marketplace where contributors earn revenue share. The post also benchmarks fast-growing peers like Cursor ($1B+ ARR, $29.3B valuation) and Lovable ($200M ARR in 8 months), and outlines conservative revenue projections reaching $50 million by 2030.

This is an English translation/structured summary of a Chinese forum post introducing Deepractice, a Hong Kong startup building a general-purpose AI agent platform.

Key points

The vision: an 'AI employee' for every industry

Deepractice, founded in 2025 in Hong Kong, wants to do for AI agents what the JVM did for Java: provide a universal runtime so intelligent agents can run anywhere. Users — from e-commerce sellers to educators to developers — can build AI assistants through simple conversation, with no coding or server setup required, paying only for usage.

The post contrasts this with the success of vertical AI tools:

Cursor: grew from zero to $1 billion annual revenue in 24 months; valued at $29.3 billion by end of 2025.
Lovable: reached $200 million annual revenue in 8 months, a SaaS speed record.

What is an AI agent?

The article distinguishes chatbots (like ChatGPT) — which only answer questions, like a 'consultant' — from AI agents, which autonomously plan steps, call tools, and complete tasks, like an 'employee.'

Example: Manus, a general AI agent that autonomously browses, researches, and produces reports, drew 2 million users on its waitlist within 7 days of launch in 2025 and was later acquired by Meta for over $2 billion (deal completed end of 2025, currently under Chinese regulatory review).

Market outlook

Per MarketsandMarkets and Grand View Research:

| Year | Market size (USD) | CAGR | Main driver | |------|-------------------|------|-------------| | 2025 | $760-780M | - | Early enterprise adoption | | 2026 | $1.04-1.12B | 43-46% | Multi-agent systems and LLM integration | | 2030 | $4.83-5.26B | 43-46% | Full automation in education/services | | 2033 | $18.3B | 49.6% | Regulatory maturity and ethics focus |

Gartner predicts 40% of enterprises will deploy agentic AI by 2026 (near zero in 2024). 82% of organizations report reduced downtime through AI integration.

Product: Deepractice Cloud

Conversational creation: describe needs with ~10 tags; "Chat is all you need."
Multi-LLM support: Claude, GPT, Gemini, DeepSeek, Qwen, switchable for cost/performance.
Open-source core: PromptX (3,359+ stars), DPML (HTML-like declarative config), AgentX (next-gen runtime).
Features: long-term memory, tool integration (Office docs, APIs, external systems), Docker self-hosting.
Use cases: personalized tutoring (Baiyun University, Huawei ICT Academy), 10x software development efficiency, smart city services (Nanjing Jianye project), automated consulting reports.

Business model

Free to start, pay-as-you-go pricing, fully open source to avoid vendor lock-in. A marketplace lets contributors share agents/tools and earn an estimated 20-30% revenue share, similar to Hugging Face's community model.

| Tier | Model | Example cost | Target users | |------|-------|--------------|--------------| | Free | Usage-based | $0 initial, token-based | Individuals, testing | | Pro | Monthly/annual | $20-50/user/month | Small teams | | Enterprise | Hybrid | Custom, from $500/month | Large orgs, on-premise | | Marketplace | Commission | 25% of revenue | Contributors |

Financial projections (company estimates)

| Year | Revenue | Cost | Net profit | Assumptions | |------|---------|------|------------|-------------| | 2026 | $8M | $5M | $3M | 50K users | | 2027 | $20M | $10M | $10M | Partner expansion, break-even | | 2030 | $50M | $20M | $30M | 500K users, 60% gross margin |

Cost structure: LLM API ~40-50%, infrastructure ~20%, R&D ~30%. Low marketing spend thanks to open-source community growth.

Team

A low-profile, distributed team rooted in Hong Kong's deep-tech ecosystem, centered on 13+ active GitHub repositories. Key figure: Sean Jiang, author of the DPML whitepaper. Plans to grow to 20-30 people with emphasis on AI ethics and global diversity.

References

1. Deepractice official site: https://deepractice.ai 2. PromptX GitHub: https://github.com/Deepractice/PromptX 3. MarketsandMarkets AI Agents report: https://www.marketsandmarkets.com/Market-Reports/ai-agents-market-15761548.html 4. Cursor Series D announcement: https://cursor.com/blog/series-d 5. Manus joins Meta announcement: https://manus.im/blog/manus-joins-meta-for-next-era-of-innovation