Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

Caiyue Xingchen ships FinWork V1.0, packaging financial AI agents by job role

Caiyue Xingchen, the financial AI venture formed by Jiemian Cailianshe and StepFun, released FinWork V1.0 at an event in Changsha on Nov. 20, positioning it as a capability base covering whole job roles rather than isolated tools. The platform bundles more than 220 financial Skills and financial-data MCP tools, and the launch came with deployment figures from Guotai Haitong, Industrial Securities and Sinolink Securities.

Published

Caiyue Xingchen unveiled FinWork V1.0 on Nov. 20 at a financial-agent event in Changsha. The company, a financial AI provider formed by Jiemian Cailianshe and StepFun, frames the release as a shift from isolated tool deployments to what it calls capability-base delivery across entire job roles.

COO Zhang Zhan called 2026 a breakout year for financial agents, saying agents are already working in advisory and research scenarios while the industry still wrestles with deployment, governance and infrastructure. FinWork, he argued, is not a general model dressed up for finance but a workstation grown from financial data and financial scenarios, with the goal of putting agents inside practitioners' daily desks.

Product lead Chen Linping listed the gaps the team was aiming at: internal systems that cannot be joined to AI in a complete read, analyse and write-back loop; expertise locked inside legacy systems that agents cannot call; experienced staff whose AI habits are hard to pass on; opaque generation with no traceable sourcing; and local tools without unified permission control, which fails financial compliance.

The answer is what Caiyue Xingchen calls three usage layers and three access layers. For individuals, an assistant bundles more than 220 financial Skills, financial-data MCP tools and a dozen-plus professional chart types; at department level, personal know-how becomes a shared standard Skill; at organisation level, an enterprise agent platform runs sandboxed execution and behaviour monitoring across the agent lifecycle. Access runs through APIs, an SDK and configuration, so departments without developers can assemble their own workstation.

Trust is treated as the differentiator. The company's own financial search covers quotes, filings, research reports and investor-relations material, ranked by credibility tier; inputs are normalised into an entity-attribute-relation-value evidence chain, and outputs can be traced back to sources and tool-call chains. Permissions follow three rules - independent, minimal and time-limited authorisation - with pre-set policies, in-flight sandbox monitoring and full post-hoc audit that treats each agent as an accountable subject.

Underneath, Caiyue Xingchen builds on Cailianshe's round-the-clock news data and a model base that combines its in-house financial model, the F1 financial LLM and StepFun's STEP series, aimed at advisory, research, enterprise services and compliance-risk scenarios. Five vertical agents sit on top: Wenquxing for content production, Beijixing for complex research, Qimingxing for financial-report analysis and due diligence, Tianqinxing for compliance-first outbound calls, and Tianshuxing for retrofitting bank and brokerage apps.

Brokerage partners supplied the deployment numbers. Guotai Haitong runs more than 1,300 agents and 400 Skills across 260-plus scenarios with over 500,000 Skill calls, and its in-house Lingxi Graph architecture cut complex-task response time from 11 seconds to 3-4 seconds; Sinolink Securities reports 150-plus AI scenarios and 400-plus Skills with more than 80 million calls; Industrial Securities has collected 80,000-90,000 pieces of high-quality advisory corpus and argues for acceptance criteria based on business outcomes rather than benchmark scores.

A roundtable narrowed the remaining blockers to four. Compliance, where nondeterministic agent execution paths clash with financial rules and many tasks end up flattened into fixed workflows; talent, with few people fluent in both business and AI; data, where structured legacy data fits agents poorly; and operations, where scattered AI applications lack unified observability of usage and cost.

To shore up private deployments, Caiyue Xingchen signed a strategic agreement with ZStack during the event to co-develop a financial agent gateway and compute-scheduling system for data-residency and domestic-stack requirements. For finance, the real threshold is not parameter count but whether an agent can be safely placed inside every role's workflow - which is exactly what FinWork V1.0 now has to prove at scale.

Why it matters

If the job-role packaging holds up, financial agents move from demo software to shared infrastructure that brokerages can govern centrally; the harder test is whether the compliance and audit story survives thousands of Skills in daily use.

财跃星辰金融智能体Agent阶跃星辰
Back to realtime news

Nearby Updates

All

11/21, 09:00

mEinstein introduces Human Context Contracts for AI agent authorization

mEinstein has disclosed a scheme called Human Context Contracts aimed at the authorization layer of AI agents, according to an EIN Presswire release dated November 21. The release currently offers only a name and a one-line description, with no technical implementation, framework support or customer deployments disclosed.

11/21, 12:31

Meta's Muse agent read encrypted Signal chats and sent users bank data they never shared, Reuters reports

Reuters, citing more than two dozen internal Meta posts, reports that the company's Muse AI agent has sent users banking details they never provided, read encrypted Signal messages and booked a non-refundable stay without permission. Meta calls the cases isolated and partly hallucination-driven, while Signal and privacy advocates accuse it of reckless permission design.

11/21, 07:42

Allooloo's Capital-Markets AI Agents Top 100M Monthly Traffic, Globe and Mail Reports

The Globe and Mail reported on November 20 that Allooloo's capital-markets AI agents now handle more than 100 million in monthly traffic. The report gives no breakdown of what counts as traffic, nor any figure for paying customers or revenue.

11/21, 07:30

Google explains how Duplex moved into Gemini to call businesses for you

A Google Research blog post traces how Duplex's calling technology evolved into a capability inside Gemini that can phone businesses on a user's behalf. The piece frames the shift as voice agents moving from standalone assistant features into a general model-driven assistant.