On almost every institutional trading desk, there tends to be one person who really understands the data infrastructure, who knows the APIs, the schemas, the quirks of the analytics layers, and who spends a disproportionate amount of their time fielding ad hoc requests from colleagues who cannot access the data themselves. It is a bottleneck that has persisted for decades, and one that ExeQution Analytics CEO Cat Turley experienced first-hand across a 25-year career spanning technology and front-office roles.
Eolas, launched by ExeQution Analytics this week, is an agentic AI assistant for trading desks designed to eliminate that dependency. Built on Anthropic’s Model Context Protocol (MCP), it converts natural language queries into structured API calls against a firm’s own market and proprietary data, allowing anyone on the desk – trader, sales trader, head of trading, compliance – to interrogate the data directly.
“What we see this tool as doing is democratising access to data across the desk, so that nobody needs to learn the API or understand the schema of how data is stored,” Turley tells TradingTech Insight. “You can just ask simple questions and get access to the information you need.”
Buy Side, Sell Side, and the Bottleneck in Between
Turley says the response from both sides of the market has been strong, but for different reasons. On the sell side, electronic coverage teams are under pressure to provide analytical colour while managing large algo books. Sales traders want rigorous preparation for client meetings rather than relying on gut feel.“Sales traders love it because it enables them to get data-driven analysis for their client meetings, so that the conversations they’re having aren’t anecdotal,” she says. “They’re not based on feeling, they’re based on actual data.”
On the buy side, the appeal centres on TCA and broker panel evaluation, processes that are notoriously time-consuming when only one or two people on the desk can query the underlying data. Quant and execution analytics teams, meanwhile, welcome the relief from constant ad hoc requests.
“They see their day freeing up to work on what they were actually hired to do – the deeper thinking work – instead of being the only one who knows how to use the API and therefore having to field all these requests,” Turley says.
Constraining the Agent
The central technical claim for Eolas is hallucination resistance. Rather than allowing the LLM to generate answers directly, Eolas converts natural language queries into API calls against a controlled function library. The AI acts as a translator between human language and structured data queries, not as a computation engine.“The first time a trader gets a wrong piece of information from this tool is the last time they’ll use it,” Turley says. “When there is so much money on the line, we need to ensure the results are entirely consistent.”
Because the AI is limited to calling approved functions on approved data, it cannot fabricate an answer. The worst it can do is make an invalid API call.The API returns an error, and the model retries. Turley gives a practical example: the model requests a breakdown of client behaviour using ‘summary equals client code.’ The API rejects the ‘client code’ parameter; the model recognises ‘client name’ is the correct value, substitutes it, and retrieves the data on its second attempt.
“If we can ensure the worst thing that can happen is an error, then when you do get a result, you can trust that result,” she says. “What we didn’t anticipate was that the agentic part of it would read the error and try again on its own.”
MCP, KDB, and the Analytics Gap
Eolas’s architecture places the MCP server between the client’s AI interface (typically Claude, Gemini, or Microsoft Copilot) and ExeQution’s analytics framework, written in q and running on KDB.
“Eolas is designed as a model-agnostic tool, because the models are all evolving at different rates,” Turley says.
The KDB dependency is deliberate. ExeQution Analytics was founded five years ago on the observation that firms had built powerful KDB databases but underinvested in turning that data into actionable analytics. Quants were either having to learn q or bypassing it entirely and pulling raw data out. Eolas is the natural extension of that situation, adding a natural language layer via MCP on top of the analytics API.
On data security, Turley is emphatic: everything lives within the client’s own environment, integrated with their existing entitlements and AI strategy. ExeQution does not contribute back to the open-source MCP standard.
“We’re very aware of the concerns around data leakage and information leakage – most of our clients are large banks and large buy-side firms,” she says. “It’s designed to be fully auditable and very transparent, so that people can have confidence in the security aspect.”
A Consultancy Play, Not a Product Launch
Perhaps the most counterintuitive aspect of the Eolas launch is the commercial model. There is no licence fee. ExeQution operates as a consultancy: it sends specialists into client organisations, charges for their time, and builds a bespoke implementation tailored to each firm’s data environment, trading workflows, and AI strategy. Every deployment is different.
“We don’t want to offer something that works off the shelf, because if you make something work off the shelf, you have to make it work for the average of all situations,” Turley says. “If you’re working towards an average, you’re losing a lot of the nuance of the data.”
Clients are expected to maintain their own relationship with KX for the underlying database technology. ExeQution has a close partnership with KX, but there is no OEM arrangement. The function library that defines what Eolas can and cannot do is built jointly with each client, with the division of labour depending on internal capability.
“A lot of our work is about training up internal people to have the understanding we have,” Turley says, “so that they don’t feel handcuffed to us.”
Building Blocks, Not Black Boxes
Turley frames the value proposition in terms of acceleration rather than replacement. Firms could build this capability themselves – and many are trying – but ExeQution’s pitch is that two decades of accumulated experience with KDB analytics and trading desk workflows compresses the development timeline considerably.
“People will do everything in house and make all their mistakes for the first time, whereas I’ve made all the mistakes over 20 years and I’ve learned a lot from them,” she says. “Our model is: learn from the mistakes we’ve made and fast-track your development by taking advantage of these building blocks and our experience.”
ExeQution Analytics was launched in Australia in 2021 and now services a global client base including hedge funds, sovereign wealth funds, proprietary trading firms, asset managers, and brokers across all exchange-traded asset classes, with FX being added this year.
Subscribe to our newsletter
