Angel-Round Portfolio Company Joyskim Raises Tens of Millions of Yuan, Has Achieved Break-Even | Yunqi Partners

On September 11, Joyskim, a company backed exclusively by Yunqi Capital at the angel round, closed a Pre-A round worth tens of millions of yuan. The new funds will go mainly toward training the AIGO model, advancing the ongoing evolution of its AI FDE Model, refining its RSI closed loop, and accelerating the replication of already-validated use cases in the market.

While most enterprise AI companies are still struggling with POC conversion and renewals, Joyskim — four years old and already at breakeven — has posted a different kind of report card: nearly 100 paying customers, a CAGR of over 300% in both orders and revenue from 2023 to 2025, and first-half 2026 revenue already surpassing all of 2025.

The following is excerpted from ChinaVenture, by Zhigao Lu

On September 11, ChinaVenture learned that enterprise AI company Joyskim had closed a Pre-A round worth tens of millions of RMB, led by Infotech National Emerging Fund (Yingfu Taike), with participation from Waterwood Capital.

Founded in 2022, Joyskim takes Palantir's product as its reference point. Its focus is enterprise decision intelligence: operating in private, on-premise enterprise environments, it brings AI into data analysis, business judgment, and task execution. Put simply, they want AI to connect the entire chain — querying data across systems, making judgments, executing tasks — doing the work of that veteran in-house expert who "knows everything."

"We're not transplanting Palantir's business map to China," said Xinqi Ren, founder and CEO of Joyskim. What the company keeps is Palantir's product philosophy of using private data to power complex analysis, decision-making, and execution — but the problems it solves are Chinese enterprises' own: iterating on enterprise AI assets, building private AI infrastructure, and adapting to core business scenarios.

Peking University Alumni Founders

Benchmarking Palantir

Joyskim's founding grew out of Xinqi Ren's long-standing research into knowledge engineering, knowledge graphs, and enterprise data products — and out of a question that had nagged him for years.

Back in 2020, reviewing his projects, he noticed something: the same knowledge graph and analytics platform produced markedly different results at different clients. In his view, the variable was whether the client had genuine domain experts internally. "Experienced people can use the system to find clues, establish connections, and form judgments. If the user lacks business experience, the same platform can't deliver the same value."

Behind that judgment lay nearly two decades of professional experience. Ren studied computer science at Peking University for both his bachelor's and master's degrees, and began researching knowledge engineering — ontology, knowledge representation, reasoning — while still at PKU's AI lab. After that, he worked at Baidu on search engines, knowledge graphs, and the practical deployment of large-scale data.

In 2014, he co-founded Mininglamp Data as a technical partner, incubating and scaling the enterprise knowledge graph product SCOPA from scratch. According to Ren, SCOPA treated Palantir as a key product reference from day one, but the team didn't copy its entire business. Instead, they chose what was closer to their own capabilities and the Chinese market: private data, deep analysis, and business decision-making.

The SCOPA platform organized previously scattered enterprise data, established entities and relationships, and brought these capabilities into complex domains like finance. The resulting combination of skills — knowledge engineering, enterprise data engineering, AI technology, ToB products, and large-scale enterprise project delivery — became the foundation for Ren's next venture.

To truly close the loop from analysis to decision to action, Ren founded Joyskim in Beijing in June 2022 as an independent company. Beyond the consideration that businesses involving private data from large and mid-sized enterprises were better developed domestically over the long term, he believed solving the "expert dependency" problem required rebuilding the product and technical architecture around AI. "This kind of start-from-scratch exploration is better done inside a new company."

After going independent, Joyskim made two choices. On the technology side, rather than endlessly maintaining a traditional knowledge graph product, it plugged AI into its existing data and knowledge systems. On the market side, it moved beyond the familiar security and finance sectors, gradually shifting focus toward broader industrial scenarios like manufacturing, energy, and transportation.

The direction became clearer: continue using Palantir as a product reference, learning from its methods of using private data to support deep analysis, decisions, and action. The company set out to push what knowledge graphs had done for data organization a step further — into business execution — within Chinese enterprises' industrial and private deployment scenarios.

Ontology returned to the core of the product, tasked with organizing business objects like customers, orders, products, and equipment, along with their relationships, rules, and executable actions. "Large models know how the world works, but they don't know how this company works. Ontology is how you tell them," Ren said bluntly. The goal is to reduce the uncertainty of LLMs in core enterprise processes, letting humans and AI work off the same business structure.

The arrival of LLMs created new technical conditions for this path. In 2023, Joyskim began developing domain-specific large models. The following year, the team began experimenting with an AI FDE approach — handing part of the data processing, knowledge extraction, and scenario-building work previously done by engineers over to AI. The product was subsequently upgraded to Knora-AI.

As the product direction solidified, in December 2024 the company closed an angel round invested exclusively by Yunqi Capital. Yunqi Capital's thesis: as LLM capabilities evolve and application boundaries expand, the market has strong demand for end-to-end capability — and Joyskim has a significant edge in combining AI with data intelligence, with a complete end-to-end product and major potential in data intelligence and LLM applications.

By having LLMs take on more standardized work, Joyskim further integrated Ontology, Agent, and autonomous execution capabilities into a single product system in 2026. Yingfu Taike and Waterwood Capital came in along the company's long-built technical path, ultimately becoming shareholders.

"The key to enterprise AI is integrating private knowledge into core assets oriented toward AI, people, and the organization — and Ontology plays an irreplaceable role in that," said Wang Shu, an investor at Yingfu Taike. "The Joyskim team is among the earliest in China to propose enterprise knowledge graphs, put ontology into long-term practice, and deploy it at scale."

Letting AI Judge and Act Autonomously

The Company Has Reached Breakeven

Beyond reducing enterprise software's dependence on experts standing outside the system, Joyskim is making data, knowledge, and business capabilities freely callable by AI.

What happens when a manufacturer hits a quality problem? Customer information lives in the CRM, production batches in the MES, raw materials and suppliers in the ERP, processes and design in the PLM. In the past, people in different roles would each pull data from their own system, and business staff would piece together a judgment from experience.

Joyskim doesn't just want AI to answer questions. It first uses Ontology to organize this business information, then lets Agents handle querying, task decomposition, tool invocation, and execution — connecting processes that previously required relays across multiple people.

The effect shows up directly in time and cost. Ren gives an example: in one quality traceability workflow, cross-system queries used to take multiple people several days. After connecting Ontology and multiple Agents, a single employee can drive the entire process, with query time cut to five minutes.

On top of Ontology and enterprise data capabilities, Joyskim uses task decomposition, reasoning, and action orchestration to turn business goals into tasks Agents can execute. Beyond going deep into specific scenarios like quality management, equipment maintenance, and supply chains, they ensure data never leaves the domain and processes remain traceable — hard requirements that state-owned enterprises and leading manufacturers can't bypass.

This gives enterprises two kinds of value. One is explicit, directly measurable value: less manual labor, higher efficiency, fewer errors, shorter business cycles. The other is implicit: lower cross-department collaboration costs, the experience of veteran engineers and business experts captured in the system, and reduced knowledge loss from staff turnover.

"The real vitality of enterprise AI isn't about landing a few big customers or signing a few big orders — it's whether you can get into a company's core business, stay sticky in high-value scenarios, and keep solving complex problems," Ren said. He revealed that in a quality management scenario at a leading panel manufacturer, Knora-AI helped the client raise yield by 3 per mille — translating to gains of more than 20 million yuan.

Joyskim now has nearly 100 paying customers, mostly large enterprises. Leading private manufacturers tend to adopt software licensing and subscription models, while state-owned enterprises typically take a package of software, implementation, and services — with some implementation work handed to integrators or third parties.

"From 2023 to 2025, Joyskim's CAGR in both orders and revenue exceeded 300%. First-half 2026 revenue already surpassed all of 2025," Ren said, adding that the company has reached breakeven, expects 2026 revenue to break 100 million yuan, and aims to remain profitable.

Over the next year, they want to integrate the product more deeply with large models, improving self-iteration in real scenarios like production and R&D, gradually reducing human intervention, and achieving an RSI (recursive self-improvement) closed loop. "The short-term goal is to form closed loops in some production environments that don't require ongoing human involvement."

More than a decade ago, Ren and his team built enterprise knowledge graphs — organizing scattered data and relationships for business experts. Today, Joyskim takes a further step: putting capabilities once scattered across data, processes, rules, and expert experience into the system itself, letting AI participate in judgment and action, and turning it into a more replicable enterprise product.