LTX, the Broadridge-owned corporate bond trading platform, has begun rolling out its Agentic BondGPT capabilities to institutional clients, enabling traders to automate workflows ranging from market monitoring and order preparation to RFQ initiation and execution.
The technology, announced in June, became available to clients in mid-September following several months of testing. Around a dozen institutions were using it by mid-October, with some already deploying agents to launch trades, although LTX has yet to confirm clients executing trades autonomously.
In conversation with TradingTech Insight, Jim Kwiatkowski, CEO of LTX, the Broadridge-owned corporate bond trading platform, says the extended testing period reflected the importance of accuracy and speed in institutional trading. The company expects traders to adopt increasingly autonomous functionality as they become comfortable with the technology.Agentic BondGPT extends the natural-language capabilities of the original BondGPT platform, launched in 2023. Whereas BondGPT responds to individual queries, the agentic version can continuously monitor specified market conditions and initiate actions when those conditions are met.
From Monitoring to Execution
The platform integrates with order management systems (OMSs), allowing traders to stage orders on LTX or reflect their trading blotters in platform watchlists. Agents can then evaluate orders against market activity, available liquidity and pricing conditions.
“Imagine a user has a couple hundred items on their blotter and they’re going to have a busy morning,” says Kwiatkowski. “They set up an agent and say, for any order on my blotter smaller than $2 million, where the bond has already traded $10 million today, if LTX has liquidity in the Liquidity Cloud priced better than the last TRACE print, go ahead and execute it. Maybe they get 50 trades done and whittle away a large amount of their work so they can focus on more challenging trades.”
Traders can configure agents to generate alerts, populate trade tickets, launch RFQs or execute orders automatically, depending on their requirements and tolerance for automation.
LTX’s Liquidity Cloud provides indications of buying and selling interest that agents can compare with TRACE transaction data and other pricing sources. During a demonstration, Kwiatkowski showed agents identifying liquidity against a watchlist and triggering RFQ instructions to selected dealers.
Users can also determine whether an agent should act only once or repeatedly when its conditions are satisfied, helping prevent unintended multiple executions.
Deterministic Architecture
The system uses prebuilt components for functions including interrogating watchlists, analysing TRACE activity, checking prices, identifying counterparties and constructing trade tickets. Natural-language instructions determine how those components are combined into a workflow.
“When we build an agent, we use the generative capabilities of BondGPT to understand the intent of the user,” explains Kwiatkowski. “But we don’t use generative AI to write code to build the agent, because that would introduce risk. There’s no way to quality assure code generation at trade time. We effectively combine pre-built, pre-approved modules together based on the response we get from the large language model. Each of those modules or building blocks is quality-assured in advance like all the rest of our code to ensure that each agent works as expected.”
Clients can introduce their own data through Model Context Protocol (MCP) connectivity, although agent construction remains within the LTX environment rather than being delegated to external systems.
The platform draws on curated information sources, including security master data, regulatory filings and market data, rather than the public internet. It also provides visibility into the conditions and data used to produce results, allowing traders to examine how their instructions have been interpreted.
Controls Over Autonomous Trading
LTX has implemented operational controls that allow its client services team to intervene in automated trading activity without accessing individual client trading decisions. Client relationship teams at LTX are helping institutions configure agents and explore potential workflows.
“We’re conscious of the privacy that our clients require,” notes Kwiatkowski. “We might be helping them set up agents, but we’re not monitoring their usage. From a safeguards perspective, our client services desk knows when auto-execution trades are in the system. We’re not monitoring what those trades are, but we’re monitoring how many there are, and we’re able to stop them either one at a time or en masse.”
Kwiatkowski also sees scope for agents to take on a greater role in trade negotiation, building on the widespread use of algorithms by dealers to respond to RFQs. This could open up opportunities for more automated interaction between buy-side and sell-side trading systems.
Expanding the Scope of Automation
The current functionality centres on monitoring market conditions and initiating transactions against trader-defined criteria. More complex portfolio-level decision-making would require agents to evaluate multiple positions and potential trading opportunities together.
“Clients are already asking us for additional capabilities, as they did at the inception of BondGPT, saying ‘that’s great, can you also do this?’ For example, we have a relative-value model, and one of the requests was: evaluate my portfolio, and if you see a relative-value swap opportunity that will net me five basis points, then execute it. Fixed-income traders have way too much information to monitor. If I’m trying to do 500 trades, there’s no way I can effectively monitor all of them as a human,” observes Kwiatkowski. “I need the technology to help me.”
As institutions gain experience with agentic trading, the scope for automation is likely to expand beyond individual orders to more sophisticated portfolio-level strategies. Client demand for these capabilities is already helping shape LTX’s development roadmap, with the potential to delegate increasingly complex trading decisions to AI agents.
Subscribe to our newsletter
