The AI that remembers your product decisions — with receipts.
A role-based AI operating system for product managers: onboarding that builds a provenance-tagged brain, skills that cite it, and a loop that keeps the org honest.
Vivek Ally · vivek@brainally.io · July 2026
Every AI session starts from zero — and the fixes on the market go stale.
“Let me remind you — we’re a B2B SaaS tool for data teams. Our persona is the data lead who owns activation. We’re freemium. Our big bet right now is…”
• Context re-explained before every useful answer — outputs come back generic
• Product truth scattered across meetings, tickets and threads
• Static “context files” go stale — unmaintained, untrusted, uncited
the context tax is paid again before the first useful answer
where one PM’s product truth lives — docs, tickets, chat, calls
AI outputs carry no provenance — trust collapses at review time
Four forces converged in the last 18 months.
YC’s own RFS says the constraint has moved from model quality to domain knowledge. The bottleneck is what the AI knows about you.
Garry Tan open-sourced GBrain (MIT, Apr 2026) — ~5K stars in 24h. Mem0 raised $24M and is AWS Agent SDK’s default memory. Engines are commodity.
Glean $7.2B. Granola $125M at $1.5B — pivoting from notetaker to “enterprise AI context.” Dust $40M Series B (Sequoia).
mySecond $39/seat/mo · ChatPRD claims 100K+ PMs, bootstrapped · $49 skill packs. Willingness to pay is proven — compounding memory is not on offer.
One sitting in, the AI already knows your product — and can prove it.
A guided interview plus your existing artifacts — Notion exports, Jira boards, transcripts — become 30–50 sourced claims in one sitting. No cold start.
Core PM skills — PRD, discovery, strategy, meeting prep — load the brain first. Every output carries receipts: each claim links to its source.
Every run writes back. A weekly sweep flags drift and contradictions — like a decision a new interview just undermined.
documented > verbal > hunch > industry — in plain text, so you can override it.
Skill packs give your AI instructions. PM OS gives it evidence.
The brain is the wedge. The loop is the prize.
Brain-seeded onboarding + cited core skills. Skills free · memory metered.
Connectors: meeting transcripts first, then Jira / Linear / Slack via MCP. Retention stops depending on user effort.
Weekly strategy-drift report: what shipped vs what was decided — flagged, cited, forwardable.
Per-source permissions, whoknows, adoption dashboard. From one PM to the whole product org.
Why this order is forced: the loop needs the brain as its reference model of “what should be happening.” Brain-first is also the better GTM — day-one value, data gravity, and it avoids shipping “another dashboard.” Maps to both YC RFS: Company Brain (Blomfield) → AI Operating System (Hu).
Three compounding layers — each unlocks the next.
Context + Skills + Workflows + Automation
Knowledge / memory — provenance-tagged, compounding
Onboarding + Brain + Recursive Learning Loops
Scoped by who shares it. Every skill run draws on all four automatically — and every claim keeps its receipt.
Chat tools scope memory by thread and folder — layers you read, none of them cited. PM OS scopes memory by who shares it — and every layer cites its sources and writes back.
Bottom-up from observed pricing — cross-checked top-down. Every figure cited.
The same context + skills + workflows foundation replicates to any role — legal, sales, finance.
Knowledge-management software:
$20.2B (2024) → $62.2B (2033), 13.6% CAGR.
We earn the bigger TAM by winning one role first — investors should discount any TAM a team hasn’t earned a wedge in.
Funding verified per company; full table in sources.md.
Provenance-tagged PM memory + the PM drift loop.
Nobody ships receipts. Nobody closes the loop.
The window is narrow: Granola is building down from capture; mySecond can bolt on memory. Speed is the strategy.
The engine is free for everyone — including us. Defensibility compounds above it.
meetings · tickets · interviews
provenance-tagged · deduped · linked
PRDs & answers with receipts
more usage → more capture ↺
1 · Accumulated brain = switching cost — history, graph and provenance don’t export as value
2 · Vertical “should” encoding — PM frameworks as the drift baseline; the step YC calls genuinely unsolved
3 · Local-first trust — your repo, your git, never pushed. The anti-cloud-brain posture
4 · Org data gravity — personal brains federate into the team brain; the brain earns the OS
Escape the creator knife-fight by selling evidence, not instructions.
Content built on the receipts demo vs ChatGPT amnesia. Skills free · email captured from day one. Consulting capped at ≤20% — every engagement dogfoods the product.
Weekly drift report hand-built for 5 PM leads, then productized. Team pricing $200–500/mo. Different buyer, empty channel.
Free: 25 memories, 1 project, all skills. $29–39/mo: unlimited memory + connectors + weekly sweep. Team tier lands with the loop.
We fund the next phase with evidence, not momentum.
[Placeholders — final bios in the shared deck]
[PM operating background, domain depth, proof of shipping.]
vivek@brainally.io
Profile sought: local-first tooling, TypeScript, agent infrastructure.
One respected PM-community name materially changes the distribution math.
Why this team can win the square: encoding “what should be happening” for PMs requires PM domain depth — the one input the well-funded horizontal players don’t have.
[Placeholders — to be completed before circulation]
Raising $[ ____ ] pre-seed · [ __ ] months runway · closes at the org-brain gate (M9–12)
wedge → connectors → loop
receipts content · design partners
infra · legal · buffer
Wedge shipped + falsification gate passed · 25 paying PMs · drift loop live with 5 teams · $[ __ ]k ARR · a retention curve worth a seed round