Glimpse Markets, the buy-side data sharing network focused on the cash bond markets, has partnered with Boltzbit, the deeptech AI company, to embed live-learning, agent-based AI directly into its buy-side bond data-sharing platform, as part of a multi-phase integration programme set to begin in early 2026.
Rather than positioning AI as a downstream analytics layer, the partnership is designed to make AI a native component of Glimpse’s data fabric, operating directly within fixed income workflows where specificity, explainability, and low error tolerance are critical.
The initiative originated from Glimpse’s Buy-Side Advisory Board, which identified AI-driven insight generation as a strategic priority for the platform. “We have fifteen heads of trading and heads of fixed income on our Advisory Board who we meet with quarterly to discuss upcoming initiatives,” says Paul O’Brien, CEO of Glimpse Markets, in conversation with TradingTech Insight. “Over the past six to nine months, AI has been a consistent theme in those discussions. That naturally led to deeper conversations with Ivan and the Boltzbit team. We were impressed with what they could offer, and the partnership grew organically from there.”According to Ivan Mihov, Boltzbit’s CRO, the goal of the partnership is to give asset managers access to a controlled environment in which AI capabilities can be prototyped and refined without the long procurement and operational cycles typically associated with large-scale internal AI programmes. “There is a strong appetite to explore rapid prototyping tools that fit their processes more closely,” he points out. “As you can imagine, doing this kind of innovation within a large buy-side firm takes a long time because of procurement, operational constraints, and governance. What the buy side sees in this partnership is the ability to influence how AI technologies are applied in decision-making and alpha generation, without the heavy technical and operational overhead that typically exists. In that sense, we become something like a prototype think tank for AI solutions in the fixed income data space.”
A central design principle of the integration is that AI should augment, rather than replace, human decision-making. Early use cases are focused on automating labour-intensive analytical tasks and proactively surfacing relevant information at the point of decision. “The major shift here is moving away from static dashboards built as one-off artefacts, towards completely dynamic, almost whiteboard-like UI customisation,” says Mihov. “Where AI agents can add real value is in decision support and data retrieval. They can be fine-tuned to surface the right data at the moment it is needed. Today, traders and portfolio managers often react slowly because they need to fetch data manually. Ideally, at the point of trade, if a counterparty statistic is needed to inform who to trade with or where, an agent should proactively retrieve and present that information before the trader even initiates the search. That kind of automation around data retrieval is highly relevant. Fully autonomous execution, however, is not something the technology is ready for yet.”“Example use cases include broker reviews, pre- and post-trade analysis, counterparty scorecards, and the retrieval of contextual data such as dealer rankings ahead of a trader actively searching for it,” says O’Brien. “These tasks tend to be labour-intensive and ad hoc, often involving aggregating data from multiple sources, analysing it, and producing reports, etc. Large parts of those processes could be handled by an AI agent, which is where we see a clear opportunity.”
Boltzbit’s contribution centres on its ‘live-learning’ approach, in which models update their underlying weights in real time rather than relying solely on static, pre-trained behaviour or short-term conversational context. This enables models to adapt continuously based on new inputs and human feedback, producing highly personalised behaviour aligned with individual users’ preferences, trading styles, and query patterns. Over time, each user effectively interacts with a bespoke model rather than a generic AI assistant.
“Live learning allows a model to update itself in real time,” explains Mihov. “Traditionally, model inference and training were separate processes. You run model output in production, then periodically retrain the model offline once enough data has accumulated. With live learning, the model can update its weights from a single new data input. That means if the model encounters an edge case it hasn’t seen before, it can learn it immediately. The next time that scenario appears, it is no longer an edge case. This allows workflows where human feedback becomes part of a continuous learning loop, with the model improving in real-time, continously.”
Data governance and trust are positioned as foundational to the partnership, reflecting Glimpse’s give-to-get data-sharing model. The companies stress that the AI learns from user interactions rather than from the underlying shared data itself. Model calls are stateless, meaning no data context is retained, and Glimpse data is not used to train Boltzbit’s public models. “All data shared over the network is anonymised and platform governance is overseen by the buy-side Advisory Board,” points out O’Brien. “Asset managers have full transparency over how their data is handled and a seat at the table at every stage. Strong security protocols and data governance are critical for maintaining confidence in the network.”
Over the next 12 to 18 months, Glimpse expects success to be measured by growth in buy-side participation and data contribution, driven by stronger network effects as AI-enabled insights improve with scale. For Boltzbit, the partnership is intended to establish the joint solution as a reference point for AI prototyping in fixed income, with potential applicability across a broader ecosystem of vendors, fintechs, and trading platforms.
Subscribe to our newsletter
