Ask a plain-language question about any client and get a response built from what your team actually uploaded and recorded, with a citation for every claim, not a guess dressed up as one.
The problem
Not because the work is hard. Because nothing captured it.
Brain*form is a client knowledge base for marketing and advertising agencies that stores what your team learns about a client and answers plain-language questions with a citation back to the source.
Brain*form ingests the documents you upload, keeps a structured record for every client, and lets your team decide exactly what becomes visible to a client before anything does.
Upload a PDF, a Word document, plain text, or markdown for a client, and Brain*form breaks it into chunks it can search and cite, tagged with the language it's written in.
A question runs an exact keyword search and a similarity search over a client's documents at once, matched to the language the content is written in, not just one or the other.
Every answer names exactly which document, profile field, person, campaign, or provider grade it drew from for each claim it makes, and clicking a citation marker shows that source below the answer.
Profile fields, people on both sides of the relationship, campaigns, and provider grades all live under one client, connected in a relational graph you can click through.
Invite a client contact to a portal that reads only from views curated for that purpose: it can never see an internal note, a provider grade, or another client's record, and the client can ask their own questions and upload their own materials there.
Flipping anything from internal to client-visible shows a confirm step naming exactly what is about to become visible, and a dedicated screen lists everything currently shared with a client plus its full history.
Create a client record, fill in the profile fields that matter, and upload the documents and decks you already have for them.
Type a plain-language question about that client, and Brain*form searches its stored profile, people, campaigns, provider grades, and documents for the answer.
The answer comes back with a citation for every claim, so you can check the exact document or record it came from before you repeat it to anyone.
Flip a profile field, person, campaign, or document to client-visible through a confirm step that names what you're about to share, then invite the client to their own portal.
Document ingestion, hybrid search, cited answers, the client record and graph, the client portal, and sharing controls all work on a Brain*form-only account. Nothing here requires another module to be turned on; it works on whatever your team uploads and records itself.
Sends
client context — Coming
The creative signal
the brand roster — Coming
The intelligence signal
brand and client entities — Coming
The performance signal
client history — Coming
The execution signal
Receives
research findings — Live
The intelligence signal
approved briefs — Coming
The creative signal
results — Coming
The performance signal
who knows what — Coming
The talent signal
Without the suite
Six months into a client relationship, half of what the team learned about them lives in one person's head.
With it
keeps what the team learns about a client as a searchable record anyone can ask a question against.
This is where a client relationship stops resetting to zero every time the person who ran it leaves, and where what the agency put its most sensitive material into gets a stated, testable containment story.
A question that used to mean interrupting a colleague gets answered directly, with the source attached, before the call it's for instead of during it.
A new hire searches instead of asking, and the person training them stops being the agency's only search function.
Most agencies already tried a wiki, a shared drive, or a Notion page to hold what the team knows about a client, and it worked for about a quarter before updates stopped and the search stopped returning anything current. What actually answers a client question today is asking whoever has been at the agency the longest, until that person is unavailable, busy, or gone.
Before
Now
A wiki page nobody has opened since the person who wrote it left
A record that updates every time someone uploads a document or logs an answer
Searching chat for a decision someone remembers making
A cited answer pulled from the actual document or record
Asking whoever has been at the agency the longest
Asking Brain*form, and getting the same answer regardless of who's in the office
A client update means digging through your own internal tool for what's safe to say
A portal built from curated views, so nothing internal is one click from a client
Models
What it decides
What it does not decide
Your data
Client content used to answer a question or generate an insight reaches Anthropic's API only when someone in your organization triggers that request, never on a schedule or in the background; embedding a document for search runs inside Supabase's own infrastructure and never reaches Anthropic at all. The answer model requires the connected Anthropic account to run under a 30-day data retention setting; that is an account-level API term set by Anthropic, not a policy this product asserts or controls. On language: keyword search matches Spanish content at full strength, and every citation displays correctly regardless of language, but the model that ranks documents by meaning is English-only, so a Spanish document's similarity-based ranking is currently weaker than an equivalent English one.
Isolation
Every table starts with row level security switched on and no default access: a policy has to explicitly grant your organization's members a read or write, and that check runs on every single request, not once at login. There is no code path for one organization's session to query another organization's client data.
Access
Every record in a client's workspace defaults to internal and stays that way until someone explicitly flips it to client-visible, a change that first shows a confirm step naming exactly what is about to become visible, whether that's a profile field, a document, or a campaign. A client's own login reaches none of your organization's real tables: it reads only through a small set of curated portal views, one per record type, each built to leave out anything internal (a campaign's internal notes, a person's private notes) and to return only that one client's rows. Providers, provider grades, private notes, and the sharing log itself have no portal view at all, in any part of the product; there is no version of the client login that can reach them. When a client asks a question in their own portal, a separate external Ask function assembles the answer exclusively from those same curated views, so the internal side of the product is never loaded into that request in the first place. All of it is checked by an adversarial test suite that logs in as a client and asserts none of it is reachable, and that suite runs again before every release, not once at launch. Every time a record's visibility changes, a database trigger writes a permanent entry recording who changed what and when, and each client's workspace has a "Shared with client" screen listing exactly what's visible to them right now plus the full history of how it got that way. Signal*form staff can access any organization's Brain*form data for support, the same access model every module in the suite uses; that access sits outside the portal path entirely and does not widen what any client's own login can reach.
Export
A client-shareable report pulls from the exact same curated views the portal itself uses, so what a report shows and what the portal shows cannot drift into two different answers to "what does this client see." A separate internal report mode exists for your own team, can include internal content, and is labeled as internal on the page. There is no server-side PDF step: the report renders as print-ready HTML, and the PDF comes from the browser's own print function.
Deletion
Archiving a client hides it from your active list without deleting anything, and it already suspends that client's portal, since every portal view filters to active clients only. Deleting a client is a separate, explicit action that removes every record under it, including uploaded files in storage, immediately. There is no self-serve organization-level deletion today.
Internal content is blocked by default at the database, a client's portal reads through views that structurally cannot see it, the external question-answering feature is built from those same views, and an automated suite logs in as a client and checks that none of it leaks, before every release. That is not a policy statement; it is a test that runs, with a pass or fail result, every time something in the product changes.
Single Tool
One signal, dialed in
billed monthly
Base
The full suite, small team
billed monthly
Pro
Growing agency pace
billed monthly
Agency
Full throttle, no limits
billed monthly
A credit is a prepaid token you spend each time the platform generates something with AI, like a brief or a report, so your plan includes a monthly batch and you can buy more when you run out.
See full pricingPrices in USD. Annual billing shown per month, billed yearly. Taxes may apply by region.
Related
Radar*form's competitive and market research saves straight into a client's record today, so a finding becomes a permanent, searchable part of that client's history instead of a slide that gets lost.
Skill*form tracks who on the team knows what; a direct link from that into a client's Brain*form record is on the roadmap, not built yet, but the two solve the same problem from opposite ends, one for people, one for clients.
Spark*form turns a client conversation into a brief for the work ahead; Brain*form is where what the agency learns about that client gets kept, so it's there for the next conversation too.