Private AI for board meeting prep
AI is a reasonable help for drafting a board packet, tightening the CEO letter or cutting a long financial appendix to two pages. The open question is where the drafting runs, because a packet is the most concentrated document a company produces: unfiled quarterly numbers, an undisclosed litigation update, an unannounced compensation change, sometimes an acquisition target.
What goes into a typical packet
A board packet usually bundles a CEO or CFO letter, financial statements ahead of public release, a litigation and regulatory update, committee reports, and any resolutions up for a vote. In a quarter with something unusual happening, that might also include acquisition or divestiture discussion, a restructuring plan, or an executive change. None of it is meant to be read outside the boardroom before disclosure. Some of it, like forward guidance ahead of an earnings call, carries securities law consequences if it leaks early.
Drafting any of that material in a general-purpose AI tool means the draft, and whatever context was pasted in to produce it, passes through a third party's infrastructure under a retention policy that wasn't written with pre-disclosure financial information in mind. Most companies wouldn't email a draft board deck to an outside vendor without a signed NDA. A shared AI tool is functionally the same transfer, without the NDA.
What changes when the drafting model is self-hosted
Running the drafting assistant on a dedicated DGX Spark means the packet, the numbers and every revision go to a model on a machine assigned only to you, not to an AI vendor's API, so no vendor retention or training policy applies. A rented Spark is still a GPUwerk-run machine (its own in Czechia, or a partner's in the country the console shows), operated under its published DPA, so check that your confidentiality rules for board material allow a hosting provider before real board documents go on it. The corporate secretary or CFO's team gets a model that can turn rough talking points into a clean CEO letter or condense a 40-page appendix into an executive summary.
Open WebUI set up for the office of the CEO or corporate secretary, restricted to that small group rather than the wider finance team, is a practical way to run this. Prior packets and templates can serve as reference material, as described in the embeddings and vector search guide, so the model matches house style and formatting without that archive leaving the group allowed to see it.
What the model is useful for, and what it isn't
A drafting model helps turn notes into prose, format a packet to a consistent template, and summarize long appendices for directors who won't read forty pages before a meeting. It has no role in deciding what belongs in front of the board, whether a matter needs disclosure, or how a resolution should be worded from a legal standpoint. Those calls stay with the corporate secretary, general counsel and the executives who own the material, the same way they would with a human executive assistant.
What this doesn't solve
Self-hosting doesn't satisfy board confidentiality obligations on its own, doesn't replace a signed NDA for anyone with packet access, and doesn't decide who on staff should be allowed to draft board material in the first place. Those stay governance decisions. What it removes is one specific exposure: unfiled financials and undisclosed corporate actions sitting in an AI vendor's chat logs before the company has decided to disclose them. See pricing for what a dedicated Spark costs for an executive office.