Noettic for Curatorial · November 2026
Meeting report, build inventory and agent team
This kit is the working record of the project: where Curatorial's systems stand today, what gets built and in what order, the team of agents that will run it, and what each side does next. It is built to be read in any order, marked up with notes, and kept current as the work moves.
Feedback goes through Notes, top right: press it, click any spot on a page, and write.
Draft for discussion · Private to Curatorial and Noettic
Meeting report and build inventory
Curatorial is turning its HubSpot contacts into a research and outreach tool for placing exhibitions, and moving its core operations onto a custom AI-driven system. Work starts at both ends at once: venue intelligence on the exhibitions side, art storage on the operations side. How much of Jarvis, the FileMaker Pro system that runs operations today, survives is still open.
Curatorial's first priority: enrich the HubSpot contacts and make the database an intelligent research and outreach tool. Museums profiled by programming, mandate and fit; directors, curators and registrars verified and kept current, with their backgrounds and expertise on file; every traveling exhibition matched to the museums and the people most likely to take it.
A standalone storage and artwork-inventory system: client profiles, item-level artwork records, rack and bin locations, instant QR check-in, and itemized client sign-off on an iPad. About 60 storage clients today, room for 100.
Rebuild the quote → work order → invoice flow with quantities per line, a verified-user gate on invoice edits, an audit log and a clean mapping into QuickBooks. Then connect everything through HubSpot and an MCP connector so staff can work with the data from Claude or ChatGPT, and put the content, newsletter and ads agents to work under one orchestrator (section V).
The meeting covered two halves of the business: art services (storage, crating, framing, installation, shipping, fabrication) run through Jarvis, and traveling exhibitions plus the Hoppé archive run through FileMaker, HubSpot, SendGrid and five websites. Today the operations systems share a single link (Jarvis pushing invoice totals into QuickBooks) and everything else is re-keyed by hand.
Where to start. Two tracks open together. On the exhibitions side Noettic builds the museums database and the agents that enrich HubSpot and match exhibitions to museum partners. On the operations side it rebuilds the core brain that FileMaker provides today, beginning with storage. Not everything in this report fits in the first 90 days; section VI sets the order.
Nothing in the Build Inventory (section IV) is committed scope until the proposal is signed.
| 1 | 60 | storage clients, with room for 100 |
| 2 | 20 | traveling exhibitions, 3–4 out at a time |
| 3 | 12,000 | newsletter contacts on SendGrid |
| 4 | 5 | websites, four of them live |
| 5 | 1 | live integration in operations: Jarvis → QuickBooks, one way |
| 6 | 10–15min | load time on the iPad storage check-in |
| 7 | 73 | items in the Build Inventory, section IV |
Every system is an island. Jarvis pushes invoice totals one way into QuickBooks, HubSpot talks to Gmail, and that is the extent of the wiring. The team re-keys the rest by hand.
The core brain. Quotes, sales/work orders signed in the field, and final invoices on Curatorial templates. Crating and framing estimators live in linked FileMaker files and push PDFs back. An items tab on each work order holds artwork checklists; QR codes on storage racks check artwork in and out from an iPad. A verbal condition-report transcription was partly built.
All billing and payment capture; receives totals, tax and payments from Jarvis at the tail end of a job.
The new CRM. Contacts migrated from a standalone FileMaker contacts file, now an archive. Enrichment used to comb the list, an email-validation workflow in place, no token-metered automations switched on yet. Gmail linked for one-to-one outreach. Breeze tried; the prospecting agent and two draft ideal customer profiles under evaluation. Nonprofit discount.
Manual macro tracking: active clients, jobs completed, services included, statuses and metrics, for storage and other jobs alike.
The exhibitions newsletter to about 12,000 contacts; chosen originally for an email API Jarvis could use.
curatorial.com and .org (exhibitions, art services, advisory); Curatorial Gallery, a year old, on ArtLogic; a neglected project site; the bespoke WordPress Hoppé site; and a Hoppé estate site held back over a copyright dispute but almost ready. Google Analytics on most.
Paid search, weighted toward art services.
The Instagram link in bio, updated by hand.
Hoppé image licensing; requests often arrive via the National Portrait Gallery and are redirected there. Decision in the room: stay on PhotoShelter for hosting.
Each traveling exhibition gets its own duplicated FileMaker database with checklist, frame label, crate label, press list and condition report layouts, hosted on the local server. The siloing is deliberate and works.
Roughly 50,000–60,000 items in two databases (prints; negatives and documents) linked by serial number by an earlier consultant.
ChatGPT is used ad hoc to structure newsletters and fix grammar; prompts are not saved. Clay was being considered for prospecting.
Installation, de-installation, framing, crating, fabrication (one-off shelves, casework, museum pedestals), local transportation by mileage, national and international shipping, storage, photography (needs definition; hourly), digital printing (currently under framing; should stay separate), and collection management as an affiliated service.
About 20 traveling exhibitions, with three or four out at any time; a printed catalog published at great expense; a newsletter that goes out around openings; curators now asking for social cross-posting with host museums.
First, a HubSpot database that finds the right museums and the right people for each exhibition. Then the thread through every other request: fewer systems, no latency, records that cover the company legally, and marketing that runs without one person as the bottleneck.
Curatorial has named this its principal priority: turn the HubSpot database from a directory of contacts into a research and outreach tool that finds the right museums and the right people for each exhibition. The first four entries are its own brief.
Each museum profiled by its exhibition programming, its institutional mandate (mission, collecting focus, scale, loan practice) and its suitability for Curatorial's projects, so the database says why a museum is worth approaching, not just that it exists.
Directors, curators, registrars and the other people who say yes, with their identities and contact details checked on a schedule, departures flagged and replacements found, rather than discovered when an email bounces.
Professional biographies, curatorial history and areas of expertise for senior staff, gathered from public sources, organized against the person's record and dated, so the team knows who it is writing to.
A ranked, reasoned candidate list for every traveling exhibition: the museums whose programming fits it and, within each, the individuals whose interests and track record make them the right first call.
The 50–100 that open each year, plus leadership changes, since published directories are stale and closed.
New museum interest categories replace the 25-year-old FileMaker codes. Research and enrichment run where they are cheapest, with clean, deduplicated results written back into HubSpot.
Tens of thousands of items, once the core is in place.
Not a subscription product: client profiles with storage terms and agreements, item-level artwork records with their own metadata, bin and rack locations attached to clients, and QR check-in and check-out that responds instantly.
Scan, photograph, dictate, and get a client signature on site instead of carrying paper and scanning it later.
A pickup of ten works is signed off as ten works, not one crate. This is a liability issue; a recent dispute hinged on a global sign-off.
Once a job is signed off, keep a PDF of the itemized record and signature and compress or purge the photos (a 20 MB image becomes about 300 KB), so the database never bloats the way FileMaker did.
Quantities on line items, invoices that can be edited after creation behind a password or verified-user gate, and a log of who changed what and when.
A materials tab, custom descriptors ("museum supplies, 60 lb of sand for Spirit House artwork" instead of "miscellaneous"), PO-compliant alternate wording, and the ability for the principal or team lead to add a new service themselves.
And does not break on long client names.
Job and metrics tracking in one place instead of a hand-kept second system.
An AWS-class object store so images stop clogging the database.
Dictation transcribed, a reference image or grid to annotate, photos attached, and eventually AI image recognition of damage. Useful for storage and exhibitions alike.
An MCP connector so anyone on the team can query the system, and a two-way link with HubSpot (a new storage account in HubSpot opening a storage profile automatically, for example).
Owned tools replace paid tiers where it makes sense, while purpose-built services such as email delivery stay where the technology matters.
Volume of crawlable content, llms.txt and robots.txt on every live site, and a crawl-friendly Hoppé site.
Hoppé stories from the digitized book and archive, an article for every service, exhibition stories, curator and artist interviews, testimonials and museum references on the site.
At least two a month (services and exhibitions), on Kit instead of SendGrid, with topic tags, all twenty exhibitions listed in every exhibitions issue, subject-line A/B tests, and every issue archived on the website.
Pop-ups and forms on the sites feeding HubSpot; One Link in Bio pulling Instagram and newsletter content automatically.
Measures Google and Meta spend against actual quotes and jobs, and shifts some budget toward Instagram and Facebook, where curators look.
Shorts for the vertical Hoppé videos already produced; voice memo → blog post → newsletter → social.
Of the few comparable firms; two New York players were mentioned, ICP among them.
PhotoShelter and TikTok are locked to one person's phone or authenticator today.
Each agent on the model best suited to its job and cost, with a person at Curatorial approving anything before it is sent, an approval that relaxes as the system learns preferences.
Everything raised in the meeting that would have to be created, in six categories. Filter by category or search. The agents are in section V. Nothing here is committed scope until the proposal is signed.
Nothing in the inventory matches that. Try a system name (HubSpot, Kit, QuickBooks) or a word like "storage" or "sign-off".
One orchestrator and thirteen specialists, in three departments. Each runs on the model matched to its job, and nothing goes to a client, a venue or the public without a person at Curatorial approving and signing off.
Receives every request, routes it to the right specialist, tracks completion, and holds every invoice, email, post and newsletter until a person at Curatorial has approved it. Also the memory of the team: it learns Curatorial's preferences over time so the approval step can loosen.
Scores every museum against each traveling exhibition on its programming, its mandate and its exhibition history, then ranks the people inside it by curatorial background and expertise. Produces a reasoned candidate list per show, museums first, then the right first call at each. Curatorial's stated first priority.
Watches the world for new museums (the 50–100 a year) and for changes at existing ones: a director retires, a curator moves, a registrar's email changes. On a schedule it re-checks every decision-maker on file, flags departures and finds the replacement, and writes the result to the museums database and HubSpot. Flags, never silently overwrites.
Builds the profiles the Matchmaker reads: for each museum its programming and mandate, for each senior staff member a professional biography, curatorial history and areas of expertise from public sources, each dated and attributed. Runs outside HubSpot's metered features and writes clean, deduplicated results back into HubSpot.
Turns a dictated note and a handful of photos into records: a client profile, a storage account, item-level artwork entries with rack and bin locations, or a temporary job item list. Staff talk to it and upload images; it does the filing.
Takes a spoken condition inspection, transcribes it, places the notes on the reference image or grid, attaches photos, and files the report against the artwork. Later gains image recognition to flag damage.
Drafts quotes in the rewritten service hierarchy, pulls the right calculators (crating, framing, installation, transport), offers PO-compliant alternate wording, and flags any line that won't map cleanly into QuickBooks.
Trained on the Hoppé book, archive PDFs and articles; drafts stories, biographies, exhibition pieces and curator or artist interviews in the dry Hoppé tone for the principal's edit.
Writes the authority content for the services side: an article per service (the "how to light art in a museum" model), project stories, and the testimonials pages.
The voice-memo pipeline: someone talks on a walk, and it returns a blog post, then a newsletter blurb, then social captions from the same source.
Assembles the two monthly newsletters (services, and exhibitions with all twenty listed), suggests stories and images from the repositories, loads issues into Kit, and runs subject-line tests.
Captions, cross-posting with curators' and host museums' handles, and YouTube Shorts uploads from the existing Hoppé videos.
Keeps the news and blog section current from exhibitions and programming, archives each newsletter as a page, and maintains the AEO files (llms.txt, robots.txt) so the sites stay crawlable.
Reads Google and Meta spend against actual quotes and jobs, recommends reallocation, and writes captions and targeting for the Instagram and Facebook tests.
A request goes to the Registrar General, which hands it to the right specialist and tracks it to completion. Every invoice, email, post and newsletter is approved by a person at Curatorial before it is sent. As the team's preferences settle in, that approval gets lighter.
Phase 1 is the Registrar General, the exhibitions three (the Matchmaker, the Scout and the Prospector) and the Intake Clerk, with the Publisher and Webmaster arriving with the Kit migration. Phase 2 adds the Estimator, the Conservator and the content desk.
Two lanes open at once. Exhibitions starts with venue intelligence, finding museum partners, Curatorial's first priority. Operations starts with storage, the bounded problem both sides named. The marketing quick wins need no new database and land alongside.
Curatorial's written brief. It needs its own databases (museums, people, exhibitions) and a contacts-and-companies link to HubSpot, none of which depends on the storage core, so the two tracks run in parallel. The quick wins (Kit, subscribe forms, One Link in Bio, llms.txt, testimonials) need no new database and land alongside. The content agents follow once Curatorial has supplied the material in section VII.
Both sides named it. It is bounded (about 60 clients, one building), and it carries the liability problem the team feels most. Once it is running, the quoting rebuild sits on the same client and job records.
Prepared by Noettic for Curatorial from the October 2026 discovery meeting and the brief that followed it.
What Noettic does next, and what Curatorial sends so discovery can start. Tick items off as they happen; ticks are saved in your browser.
Discovery starts when these arrive.
Raised but not settled, with Noettic's recommendation where it has one. Several of these shape the scope of the first build, so they are worth deciding before Phase 1 is approved.
| 1 | Jarvis end state |
| 2 | Scope boundary with HubSpot |
| 3 | One HubSpot portal or two |
| 8 | Email platform |
| 9 | Prospecting tools |
| 10 | Sources and handling of staff data |
Eight sections, read in any order. Everything you can click is listed here, and the Notes tool is how you send feedback back to Noettic without an email thread.
The contents block at the top of every page is the navigation: pick a section and it opens in place. The roman numerals in the running head do the same thing from anywhere on the page, and the title in the running head brings you back to the cover. The address bar updates as you move, so a link copied from it opens on that section.
In Today, switch between the systems as they are now and as proposed after Phase 3, then click any system to read what it does and where it hurts. In Phasing, click a phase to see what it contains. Figures wider than a phone scroll sideways.
Seventy-three items in six categories. Filter with the category chips, or type a system name or a word like "storage" into the search box; the count updates as you go.
Next Steps has two lists with tick boxes. Ticks are saved in your own browser, so use them to track what you have sent; they are not visible to Noettic.
Everyone with the link sees the same notes, including Noettic; only the people on the review list can add to them. In the panel you can Reply to a note, mark it Resolved when it is handled, or press Go to to jump to its pin. Press Notes again to hide the pins and read clean.
The round button at the right of the running head switches the page between light and dark. It follows your device setting until you choose.
The link is private to Curatorial and Noettic; send it to anyone on the team who needs it. Reading needs no sign-in. Leaving notes does, and only the people on the review list can; ask Noettic to add someone.
This page is updated as the project moves, so the link always shows the latest version. The version and date are in the footer of every page.
| I | Overview the three outcomes and the facts |
| II | Today the systems map and the pain points |
| III | Outcomes what Curatorial wants, in 28 entries |
| IV | Build Inventory the 73 things to be built |
| V | The Agent Team the orchestrator and thirteen specialists |
| VI | Phasing the order of work |
| VII | Next Steps what each side does now |
| VIII | Open Decisions what still needs an answer |