KX unveils agentic AI blueprints for capital markets

KX has launched two agentic AI blueprints for capital markets: an AI Research Assistant and Trading Signal Agents, built with Nvidia software and KX's KDB-X data platform.

The products are aimed at research and trading teams that handle fast-moving market data and large volumes of documents. KX is positioning the blueprints for regulated environments that require audit trails, governance controls, and repeatable workflows. It also says the systems deliver sub-second response times and preserve temporal accuracy by aligning outputs to event time.

Agentic AI is drawing interest from technology suppliers and financial firms that want systems able to break tasks into steps and act across data sources. In capital markets, those approaches face constraints around latency and oversight. Trading desks and research teams also need tools that track when information became available, given the risk of using data that was not known at the time a decision was made.

Research workflows

KX's AI Research Assistant blueprint supports retrieval, summarisation, and synthesis across structured market data, unstructured content, and proprietary documents. Typical sources include earnings information, regulatory filings, and internal research notes. The goal is to reduce the time analysts spend locating and compiling information across formats.

The Trading Signal Agents blueprint focuses on identifying and monitoring potential signals in real time. KX says it uses multimodal information aligned to market time. Outputs are governed and can be assessed before feeding into trading and risk processes.

Both blueprints sit on what KX and Nvidia describe as the Nvidia AI Factory stack, using Nvidia accelerated computing and Nvidia AI Enterprise software. Listed components include Nvidia NeMo Retriever, Nvidia Nemotron embedding models, and Nvidia NIM microservices.

Data and time

KX has connected KDB.AI to Nvidia cuVS for GPU-based vector indexing and search. The platform also uses KDB-X, which KX describes as its time-series-native analytics layer. The design reflects the importance of sequencing and timing in market data analysis, where a signal's value can depend on its place in an event stream.

KX calls the approach "temporal AI," saying the system aligns data to event time and computes point-in-time-correct context. This structure is intended to support auditable workflows by allowing users to reproduce an output and verify which data points were available at the time.

Capital markets firms have adopted vector databases and embedding models to improve document search and chat-style interfaces. Many also use retrieval-augmented generation to keep outputs grounded in source material. Agent-based designs add another layer by allowing software to plan and coordinate tasks, increasing the need for guardrails and oversight.

"Capital markets are facing an alpha paradox where more data and more AI can make it harder to find durable signal," said Ashok Reddy, Chief Executive Officer, KX.

Reddy added: "The signal-to-noise ratio is collapsing. When signals decay before validation, and governance and economics can't keep up, many AI approaches break down. KX is solving these problems with Temporal AI systems that turn real-time, multi-modal data into research and signal intelligence that teams can operationalize with speed, governance, and control."

Nvidia described the partnership as a way to combine its compute and software stack with KX's time-series focus in financial markets.

"To extract durable alpha in today's hyper-competitive markets, financial institutions need unprecedented computational capacity," said Ioana Boier, Global Head of Capital Markets Strategy, Nvidia.

Boier added: "Our collaboration with KX combines NVIDIA AI Factory stack with KX's deep expertise in time-series data, enabling trading and research teams to overcome the traditional speed-intelligence trade-off and deploy complex models without sacrificing real-time execution at scale."

RBC proof of concept

KX also highlighted a production-focused proof of concept with RBC Capital Markets. The project produced an internal tool called Aiden Quick Takes, built using 14 specialised agents. KX says it compressed research cycles from hours to minutes across RBC's capital markets organisation.

The work also tested retrieval and search across structured market data, unstructured narrative, and proprietary documents used in earnings and research decisions. KX linked the improvement to integrations with Nvidia AI Enterprise software.

"We wouldn't be able to do any of this without RBC Borealis and our strategic partnerships," said Bobby Grubert, Head of AI and Digital Innovation, RBC Capital Markets.

Grubert added: "We're prioritizing industry leaders that we go deep with in terms of strategy - NVIDIA and KX being two of those firms - as we go all in to scale up and out across capital markets."

KX plans live demonstrations of both blueprints at Nvidia's GTC event and will share further detail on research and trading use cases as it prepares the debut of KDB-X as the next generation of its kdb+ ecosystem.