OPENREGULATORY
Sign in Get started
The AI Assistant
AI in Formwork Published on September 08, 2026

The AI Assistant

Chat with your QMS: what the assistant can see, what it can do, what it refuses, and how to ask well.

The AI Assistant is a chat inside Formwork that knows your company's QMS and technical documentation. Ask it which CAPAs are open, what your supplier evaluation procedure says, or whether a hazard has a risk control, and it looks it up and answers with links to the items. Ask it to draft a record or a set of hazards and it creates them for you to review.

You find it under AI Assistant in the top bar. Every conversation is yours alone: colleagues cannot see your chats, and you cannot see theirs. Chats stay until you delete them.

What it can see

The assistant works as you. It sees exactly what you can open in Formwork, and nothing else:

  • QMS documents and records, including their content and version history. Uploaded files (PDF, Word, Excel) and synced Google Docs are found by search, but the assistant cannot read their content; convert an uploaded file to a native document if you want the assistant to work with its text.
  • Technical documentation of every product and release: stakeholders, user needs, requirements, tests and test runs, hazards, risk table, risk controls, failure modes, change requests, incidents.
  • Quality processes: CAPAs and their actions, non-conformities, incidents, suppliers, trainings and training status, tasks.
  • Reviews, comments and activity history of documents and records.
  • This manual and OpenRegulatory's public templates.

It has no access to the internet, to your files, to other companies' data, or to anything in a company you are not a member of. It always works in the company you are currently viewing.

What it can do

  • Answer questions by looking things up. It cites items by their identifier (HAZ-7, CAPA-3, a document title) and links to them.
  • Search, by title across everything and by content inside documents and records, including older versions.
  • Explain relationships: which requirements a user need drives, which tests verify a requirement, which risk controls mitigate a hazard.
  • Draft new items: documents and records, hazards, requirements, tests, tasks and more. Documents, records and design and risk items it creates are drafts with you as the author; tasks it creates are live immediately, like any task you create yourself. See Drafting Documents & Records with AI.
  • Edit drafts: change the title, description or content of anything that is still a draft and that you may edit.
  • Send drafts for review to the roles you name.
  • Run a release readiness check on a product release, the same check as on the release page, and report what is missing.
  • Create tasks for open points that need a human decision, linked to the item they are about.

What it never does

  • It never approves, releases or rejects anything, never archives, restores, deletes or moves anything. Those actions stay with you in Formwork.
  • It never edits approved or released items. If you ask it to change a released document, it creates a new draft version instead, exactly as you would.
  • It does not answer how-to questions about Formwork beyond what this manual covers, and it says so when the manual has nothing on a topic.
  • It cannot upload files. Attach files to a document or record in Formwork itself, or through MCP.

Getting good answers

  • Name things the way Formwork does. "Open CAPAs", "user needs of the infusion pump", "the software requirements in release 1.2". The assistant uses the same terms as the app and understands the product-specific ones (for a software product, "software requirement" and "design input" are the same thing).
  • Search with short, distinctive words. Content search matches literal text, so "supplier evaluat" finds more than "how do we evaluate our suppliers". If the first search comes up empty, the assistant tries a shorter stem itself.
  • Give it the material. "Draft a record for today's management review; here are my notes: ..." works far better than "write a management review".
  • Ask for the check before the change. "What would you change in DI-12?" lets you approve the plan before the assistant edits anything.
  • Watch the progress line. While the assistant works, the composer shows what it is doing ("Looking up hazards...", "Drafting a document..."). Long answers stream in as they are written.

Examples

  • "Which trainings are overdue, and for whom?"
  • "Summarize what our change control procedure says about significant changes."
  • "List the hazards in release 2.0 that have no risk control yet."
  • "Create a task for each unresolved comment thread on the risk management plan."
  • "Draft a non-conformity for the labeling error we found in batch 42; severity major."
  • "Is release 1.3.0 of the app ready for release? Run the readiness check."

Good to know

  • Each message you send counts as one prompt against your company's weekly budget; the tool calls the assistant makes to answer do not. See AI Usage & Limits.
  • A company admin has to switch the assistant on once, under Settings → AI usage. Until then the AI Assistant page says so.
  • "AI can make mistakes." Treat answers as a well-informed colleague's summary, not as a substitute for reading the document. Reviewable content the assistant creates goes through your normal review; tasks and comments are live as soon as they are created.
  • If you'd rather use Claude, ChatGPT or Codex with your Formwork data, connect them via MCP: they share the same data access and the same as-you permissions, in the tool you already use. MCP can also upload files, and it does not use the weekly prompt budget. See Connect Your Own AI via MCP.