The Pulse
V7 turns company files into agent memory
OpenAI says V7’s Context Graph lets AI agents organize millions of company files into searchable records of entities, relationships, and cited evidence. The finance-focused startup reports faster deal screening, lower document costs, and 89
AI.info Team ·
V7 is building an AI system that treats a company’s accumulated files less like a pile of documents and more like a working record of how the business operates. In a profile published by OpenAI on September 21, the finance-focused startup describes V7 Go, an agent platform that extracts entities, relationships, facts, and evidence from company repositories and organizes them into a searchable Context Graph.
The approach is aimed at workflows in finance, insurance, and real estate, where the answer to a question may be spread across reports, spreadsheets, emails, data rooms, and internal systems. V7 says its agents can use that structured information to complete workflows involving dozens of steps while preserving an audit trail of the decisions and sources involved. OpenAI’s account of V7 identifies the company as a startup serving customers in Europe and the United Kingdom.
V7’s Context Graph stores more than document text
V7 Go connects to repositories including SharePoint and Google Drive, then scans incoming material for companies, funds, people, metrics, and other entities defined in an ontology. Each fact is linked to an existing or newly created record, with evidence preserved back to the original source. When the graph does not contain enough information, the system can search the underlying files through retrieval-augmented generation.
That structure is designed to reduce the repeated searching that agents often perform when they must reconstruct business context from scratch. V7 says the graph is an order of magnitude cheaper and faster to traverse than long-context approaches, although OpenAI’s profile does not provide the underlying cost or latency measurements behind that comparison.
The system also gives agents a longer memory. Recent exchanges remain in active model context, while older material stays in the graph and can be retrieved when needed. The arrangement lets a workflow use a firm’s history without requiring the model to relearn the same relationships on every request.
“To solve hard enterprise use cases across finance and insurance, AI needs to learn how your business operates just as well as it learned from the Internet.”
Alberto Rizzoli, co-founder and CEO at V7
Finance workflows show the system’s intended payoff
V7’s examples center on document-heavy work that normally requires specialists to compare records, extract terms, and check inconsistencies. In one workflow, V7 Go processes a Confidential Information Memorandum, pulls out financial details, deal terms, management information, and risk fields, then produces a screening note with citations.
According to figures in OpenAI’s profile, asset managers using the system screen deals 21 times faster, reducing a full-day process to 15 minutes. A financial services team cut review time from more than 100 hours to under 10 hours, with V7 attributing $12,000 in expert-cost savings to each task. Insurance teams reduced claims-processing errors by 13.5% against a manual baseline after giving agents access to historical claims and existing policies.
Those results come from V7’s own customer and workflow measurements, not an independent evaluation. The company says its agents complete workflows spanning 50 to 100 steps in minutes and reach 99.9% accuracy, but the profile does not define the full test methodology for that figure.
OpenAI models handle extraction, reasoning, and tool use
V7 uses different OpenAI models for different parts of the system. GPT-5.6 Luna handles high-volume extraction and other structured tasks, while GPT-5.6 Terra and Sol support chat, tool use, and more demanding workflow steps. V7 has begun testing GPT-6 Astra on the hardest queries against its Context Graph.
For that test, V7 created a more difficult set of graph questions using messy data drawn from thousands of documents. The company reports that GPT-5.6 Sol scored 78% on the very-hard tier, while GPT-6 Astra reached 89%. Both models scored close to 100% on the easy, medium, and hard tiers, according to the OpenAI profile.
V7 also reports a reduction in tool-call errors after moving from GPT-5.5 to GPT-5.6 Sol. The error rate fell from 2.7% to 0.2% in its Context Graph benchmark, a result the company attributes to the model’s ability to follow multi-step instructions and interpret tool results.
“We chose OpenAI as our default because it performs best on the multi-step tool workflows V7 Go depends on. In our Context Graph benchmark, GPT-5.6 Sol reduced the tool-call error rate from 2.7% with GPT-5.5, to 0.2%.”
Simon Edwardsson, co-founder and CTO at V7
MCP connects the memory layer to ChatGPT and Codex
V7 exposes Context Graph ingestion and queries through its MCP server, allowing customers to use the system from ChatGPT and other compatible clients. Developers can also create V7 Go workflows through MCP in Codex. The company says those integrations, combined with simpler workflow design, cut the time needed to build a medium-length workflow from about one hour to roughly 20 minutes.
V7 has also moved document-heavy workloads from OpenAI’s Chat Completions API to the Responses API. In its testing, the change reduced token use by about 5% for some PDF-heavy workflows and improved caching reliability. GPT-5.6 Luna lowered cost per document by 78% compared with GPT-5.4 mini in another V7 measurement.
V7 wants the graph to flag changes before people ask
The next stage described by V7 is a more proactive memory system. The company is working on workflows that start when facts in the Context Graph change, identify inconsistencies, and show people which earlier analyses may need review.
One example is a restated fund report that could trigger a warning about downstream work still relying on older figures. That would shift the graph from a store of retrievable company knowledge into a monitoring layer for decisions already made.
“Our goal is to help enterprises re-tool for the age of AI, with workflows that solve mission critical tasks, and memory that outperforms us humans,” Rizzoli says in OpenAI’s profile. For now, the clearest evidence of V7’s approach is narrower: a graph that links business facts to their source documents, and a set of customer-reported results built around finance and insurance workflows.