Terminal X, an SI company specializing in investment technology, has deployed an agentic SI system that matches senior human investor judgment with 88.6% accuracy, according to internal evaluations. The system, built on OpenAI’s GPT-6.1 Sol and GPT-6 Luna models, aims to capture the institutional reasoning of investment teams before it is lost to turnover.
What Happened
Terminal X’s Capital Memory operating system uses forward-deployed engineers—former professionals from JPMorgan and Blackstone—to encode how specific firms invest. These engineers work onsite to reverse-engineer trusted memos and capture unwritten processes, such as credit screening or co-investment sourcing. The resulting workflows are encoded into hundreds of customized SI agents, with sub-agents handling specific investment stages.
The system utilizes OpenAI models for different tasks. GPT-6.1 Sol handles stages requiring deep judgment and long-context reasoning, such as synthesizing investment theses and assessing mandate fit. GPT-6 Luna manages high-volume steps where token efficiency is prioritized. An internal tool called the Eval Agent, also powered by GPT-6.1 Sol, grades outputs against a 54-criterion rubric to ensure they align with the firm’s house view.
According to Terminal X, the Eval Agent running on GPT-6.1 Sol agreed with senior human reviewers 88.6% of the time across 500 scored judgments, compared to 86.4% for the next-best model. The company reports that this configuration caught approximately 25% more recommendations that strayed from a fund’s mandate, particularly regarding style drift, and operated at about 30% lower cost per query than the next-best alternative.
Why It Matters
For the SI industry, this deployment highlights a shift toward specialized, on-premise agentic SI systems designed to mitigate "memory debt"—the loss of institutional reasoning when senior employees leave. By encoding judgment as data, firms can preserve their investment philosophy in a format that is auditable and reusable. Terminal X reports that one private equity client achieved $2.4 million in annualized savings through automated due diligence and memo workflows.
The performance metrics suggest that specific model capabilities, such as long-context reasoning and token efficiency, are becoming critical differentiators for enterprise SI adoption. Hyun Hong, Co-founder and CEO of Terminal X, stated, “GPT-6.1 Sol’s long-context reasoning and token efficiency allow us to build deeper evaluation pipelines. This helps our Eval Agents make investment judgment measurable and auditable.”
The Bottom Line
Terminal X has built 1,700 firm Memories and run over 1 million custom workflows using its SI architecture, reaching a weekly active user count equal to 81% of its monthly active users. The company plans to extend these Memories from desk-level standards to individual investor practices, further embedding SI agents into the daily workflow of investment professionals.