V7 Equips Its AI Agents with Corporate Memory

V7 highlights V7 Go, an agentic platform that transforms heterogeneous repositories into a context graph queryable by agents. The company claims speed and accuracy gains on critical flows in finance and insurance, relying on OpenAI models to drive multi-step sequences with complete traceability.

Reported Gains on Critical Flows in Minutes

V7 reports that asset managers are now filtering transactions 21 times faster, reducing a one-day process to just 15 minutes. A financial services team reportedly cut an examination from over 100 hours to less than 10, saving $12,000 per task in expert costs. In insurance, the company mentions a 13.5% reduction in claims processing errors compared to a manual benchmark, after providing agents with historical knowledge of claims and policies. V7 also claims that agents complete flows of 50 to 100 steps in just a few minutes with 99.9% accuracy. The platform emphasizes maintaining a verifiable traceability of every decision and execution.

Structuring Business Context in a Queryable Graph

V7 describes the gap between the reasoning capabilities of models and the lack of spontaneous understanding of business context. This context is scattered across documents, data rooms, spreadsheets, emails, and internal tools, making it poorly accessible to agents. According to the publisher, this deficiency forces agents to rediscover information with each query, multiplying costly searches and missing essential links. To address this, V7 Go connects to repositories like SharePoint and Google Drive, extracts entities, relationships, facts, attributes, and metrics, and then fills a Context Graph that is claimed to be an order of magnitude cheaper and faster than long-context approaches. This graph provides a structured and up-to-date record, directly queryable, where each new file enriches existing records while preserving cited evidence. If the graph proves insufficient, a search in the underlying documents via RAG remains possible. For extended executions, the recent context is retained in the model's active context, and the history is archived in the graph for on-demand retrieval. V7 asserts that this source anchoring maintains the relevance of complex flows for each business.

Document Use Cases and Internal Reference

The company positions its use cases in private equity transaction analysis and insurance underwriting management. During a presentation, a flow extracted elements from a Confidential Information Memorandum, gathering financial data, transaction terms, and information about the management team, identifying risk categories, and then producing a filtering note. V7 specifies that it measured the contribution of structuring on HERB, an internal repository designed to connect scattered data within the organization: its retrieval tool alone reportedly exceeded the official benchmark of 69% and reduced hallucinations on unanswered queries by 38%.

Level Orchestration and Model Selection

For extended action sequences prone to errors, V7 offers supervision combining deterministic code, delegation to smaller models, file creation, and integrations, while ensuring an audit trail with each launch. All steps of an AI-generated flow are organized into three categories — fast, medium, intelligent — allowing for the assignment of the appropriate engine. GPT-5.6 Luna handles structured extraction and high volumes, while GPT-5.6 Terra or Sol manage discussions, agent driving, and more demanding reasoning and tool steps. V7 states it has begun using GPT-6 Astra for the most challenging queries of the Context Graph, such as financial analysis across thousands of documents.

Exposing Enterprise Memory to ChatGPT and Codex via MCP

V7 Go exposes the ingestion and querying of the Context Graph through an MCP server, enabling use from ChatGPT and other compatible clients. Clients can also create V7 Go flows via MCP in Codex. According to the company, combined with simplified design, this integration has reduced the time to create an average-length flow from about one hour to approximately 20 minutes.

V7 Bets on OpenAI and Envisions Proactive Memory

V7 explains that it continuously evaluates models based on a battery covering citation accuracy, extraction quality across hundreds of document types, correctness, instruction tracking, latency, cost, and real-world scenarios. On these criteria, the company claims that OpenAI outperforms most alternatives, noting that its capacity increase requests have been approved and implemented within hours, compared to weeks with others. The publisher states it is working to make shared memory proactive, with flows triggered by factual changes, reporting inconsistencies, and indicating analyses to review, illustrated by the example of a revised fund report. Rizzoli asserts the goal of helping businesses retool for the AI era and maintains that, in finance, value will come from better context rather than a greater number of agents. V7 reiterates its focus on critical flows and a queryable memory, in a framework where the demand for accuracy is presented as non-negotiable, following a foundation established in 2018 by Rizzoli and Edwardsson.