V7 aims to give AI agents business context from the scattered documents and data of finance, insurance, and real estate companies. Founded by Rizzoli and Edwardsson in 2018, the company’s V7 Go platform brings together information from millions of files in a structure called Context Graph. This structure links entities, relationships, and source evidence, enabling agents to query information directly instead of searching for it again.
V7 Go uses GPT‑5.6 Luna for structured information extraction, and GPT‑5.6 Terra and Sol for complex instructions, reasoning, and tool use. GPT‑6 Astra is also being tested for the most challenging Context Graph queries. The company says that agents complete 50–100-step workflows within minutes and maintain an auditable record for every decision. The system extracts companies, funds, people, relationships, and financial metrics from repositories such as SharePoint and Google Drive, matching them with evidence in the source documents; if there is not enough information in the graph, it continues searching the documents with RAG.
In V7’s HERB test, the access-only system delivered 69% better results than the official baseline model and reduced hallucinations in unanswered questions by 38%. The reported effects of Context Graph on the company’s clients are as follows:
- Asset managers accelerated deal screening by 21x, reducing a full-day process to 15 minutes.
- A financial services team reduced review time from more than 100 hours to less than 10 hours and saved $12,000 in expert costs per task.
- Insurance teams reduced claims-processing errors by 13.5% compared with the manual baseline after making historical claims and policy information available to agents.
Workflows in V7 Go manage code, handoffs to smaller models, file generation, and external integrations together. V7’s tests cover citation accuracy, document extraction, response accuracy, instruction following, latency, and cost. GPT-5.6 Luna reduced the per-document cost by 78% compared with GPT-5.4 mini. Switching to the Responses API reduced token usage by approximately 5% in some PDF-heavy processes.
On the most difficult graph queries, GPT-5.6 Sol achieved 78% accuracy, while GPT-6 Astra achieved 89%; both models reached approximately 100% on the easy, medium, and difficult levels. V7 Go’s MCP support enables Context Graph and workflows to be used from ChatGPT, Codex, and compatible clients; the time required to create a medium-length workflow fell from approximately one hour to 20 minutes. V7 is also developing more proactive workflows that are triggered by changes in the graph and flag inconsistencies.
Why it matters
This development aims to transform AI agents at financial, insurance, and real estate companies from tools that merely search documents into systems capable of using the relationships and supporting evidence between documents. This could change how human experts access information, particularly in document-intensive processes such as deal review, financial assessment, and claims processing. The company’s customer examples show that this can affect processing times, expert costs, and error rates; however, these results are based on V7’s own reports. GPT-6 Astra’s performance on the most difficult queries should likewise be evaluated within the scope of V7’s test conditions. The open question is whether the Context Graph approach can deliver the same accuracy and cost results across different organizations’ document structures and workflows.
Term: agent
An AI agent is software that calls tools and carries out multi-step tasks to achieve a goal, rather than generating a single response.