Legal

How VektorForgeAI uses AI

Last updated: · Effective

Our Platforms use artificial intelligence to help you plan and deliver your work faster. This page explains which AI features each Platform has, what information they use, where it goes, and what you should keep in mind. It forms part of our Privacy Policy, and words such as Platform, Your Content, AI features, AI provider, AI output, external AI assistant and plan limits have the meanings in the Terms of Service.

AI features on each Platform

Product Architect

  • AI drafts and recommendations across your product documents: strategy, vision, requirements, roadmaps, sprints, business cases, mission and charter documents, RACI charts, metrics, release reviews, launch readiness checks, retirement plans, decision records, architecture diagrams and prioritization.
  • Arc, the AI coach, a chat assistant for your product.
  • File import, which reads documents and images you upload (such as PDF, Word, PowerPoint, Excel, CSV and image files) to pull out information.
  • Feedback analysis, which summarizes customer feedback you enter.
  • Market Research and Deep Research, AI research features that can search and read public web pages to answer research questions, with links to their sources.

Program Forgemaster

  • Flint, the AI coach, a chat assistant for your program that answers questions from your program’s records and can propose changes to them, and which also drafts a program charter, suggests benefits and dependencies, drafts a program roadmap with proposed dates, reviews your program’s Connect and Deliver stages, turns meeting notes into draft entries for your program’s logs, drafts whole status reports (you set every red, amber or green status yourself) and the words of closure reports, drafts a recovery plan to get your program back to green, with the effect each step is expected to have, suggests lessons learned, tells you when a record needs review because something it is built from changed and suggests updates to it, and can walk you through setting up a program step by step. Flint saves its drafts and suggested changes in your program, labeled as Flint drafts: they count for nothing and change nothing until you approve them, and you can edit or discard them. Flint can also suggest short notes to remember about your program; it keeps only the ones you accept, and you can edit them or tell it to forget them at any time.
  • Who uses Flint on a shared program. Only the program’s owner and its program and project managers. Sponsors and viewers don’t use Flint and never see Flint’s drafts, Flint conversations or what Flint remembers. Each person who uses Flint has their own monthly allowance and daily cap, and their Flint conversations are visible only to them.

We will describe any AI features on Systems Alloysmith, when it launches, on this page before they are available.

On every Platform, AI features run only when you, or someone you share with, uses one of them. We don’t send Your Content to AI in the background.

Which AI we use

All AI features use Claude models from Anthropic, our AI provider. Requests go from our servers to Anthropic; your browser does not connect to Anthropic directly. We will update this page before we use a different AI provider.

What information is sent

For each request, we send Anthropic only what that feature needs:

  • your prompt or question, and recent messages in the same conversation;
  • the relevant information from your work, such as the document you are working on, your product’s market research or your program’s records;
  • files or images you attach to the request;
  • for feedback analysis, the customer feedback you have entered, and for Flint’s meeting notes and lessons learned features, the notes you enter;
  • in Program Forgemaster, what Flint remembers about your program (the notes you chose to keep), and, for Flint’s suggested updates, the earlier and current versions of the records that changed.

AI research features also send search queries based on your question to Anthropic’s web search tool, which retrieves public web pages.

How your information is protected

  • No training on Your Content. We do not use Your Content to train AI models, and Anthropic does not use data sent through its commercial API to train its models.
  • Limited retention by Anthropic. Anthropic deletes request data within 30 days, unless a request is flagged for violating Anthropic’s usage policies, in which case it may be kept for up to two years.
  • Safety checks without your text (Product Architect and Program Forgemaster). Our servers check AI requests and AI answers (and, in Product Architect, content written through an API or sent to a metric’s intake address) for known manipulation patterns, such as instructions hidden in a document. When one matches, we record which check matched, where it was found (for example, in a request or in an answer), how many times, the account (and the API key, if one was used), the AI feature, the item and when, never the text, and keep that record for 12 months for security. If the same match is found again in the same item’s content within 24 hours, we record it only once. Text inside images, and in some PDFs, can’t be checked this way; for those we check the AI’s answer.
  • AI usage records without your text. To apply plan limits and control costs, we record which AI feature was used, the model, the amount of text processed and the number of web searches, but not the text of your prompts.

What we keep

  • AI output you save becomes part of Your Content, with its version history kept as described in our Privacy Policy.
  • AI coach conversations are saved with your work so you can pick up where you left off: in Product Architect with your product, and in Program Forgemaster with your program. In Product Architect, a copy of the current conversation is also kept in your browser’s local storage.
  • Research reports, and the reasoning behind AI recommendations, are saved with your work.
  • In Program Forgemaster, Flint’s drafts and suggested changes are saved with your program until you approve or discard them, and what Flint remembers about a program is saved with it until you change it or choose “Forget all”, which deletes it without keeping a copy.

Deleting an item deletes its AI conversations, drafts, research and what the AI coach remembers about it, along with everything else in it. In Program Forgemaster, you can also delete a Flint conversation on its own.

AI can be wrong

AI output can be inaccurate, incomplete, out of date or biased, and it can sound confident when it is wrong. Research answers depend on the web pages found. That’s why AI output is labeled as a draft, and why we show the reasoning and sources behind many recommendations. Review AI output before you rely on it or share it. It is not legal, financial, tax or other professional advice, and our Acceptable Use Policy does not allow it to be the only basis for decisions with legal or similarly significant effects on people.

Your choices

  • You decide when to use AI features; the rest of each Platform works without them.
  • Don’t put information into the Platforms that you don’t want processed by AI, and never upload sensitive information such as payment card numbers, government ID numbers or health information (our Acceptable Use Policy lists it all).
  • AI features are subject to plan limits, including daily AI usage caps, and we monitor usage to prevent runaway costs.
  • Your external AI assistant. On Platforms and plans that include the API, you can connect an external AI assistant, such as Claude or Cursor, to your work through our Model Context Protocol (MCP) server, using an API key. The assistant then reads and changes your documents on your instructions. It runs on its own provider’s systems, not ours, and its provider processes what it reads under its own terms, even if that provider is Anthropic. Our AI plan limits don’t apply to it, but API request rates do.

Questions

Email info@vektorforgeai.com.