For private equity firms · Formerly Horison
Your firm has looked at this company before.
Seisa reads the email, data rooms, CRM and call notes your firm already produces, resolves them into one record per company, person and deal, and answers with the document and the page.
Your team and your AI agents reason from the firm’s own history instead of from scratch.
The problem
A firm’s only durable edge is its own deal history and judgment.
Today it lives in drives, inboxes, CRMs and people’s heads. The 2019 pass memo, the expert call that flagged the distributor risk, the partner who knows the CFO. Each is somewhere, none of it is in the room when the next deal starts.
When people move on, it leaves with them. Every process starts from zero.
Pass memo
Project Aster — recommendation
Held at 9.5x against an 11.2x ask.
Revenue concentration in the top three accounts.
Nov 2021 · p. 3
a partner’s drive
Email thread
Intro — Northwind CFO office
Tom Whitaker to A. Ferreira
Happy to make the introduction.
4 messages · 2019
an inbox nobody searches
Expert call
Former distribution lead
00:41:12the distributor risk is the one nobody prices
00:41:30it took two years to show up
62 min · unfiled
a call-notes PDF
How it works
We read what the firm already produces.
Seisa connects to the systems you have: data rooms, email, CRM, Slack, Snowflake, call notes. Read-only. No migration, no new workflow, no one re-entering anything.
Connected sources
Outlookemail
Gmailemail
Google Drivedocuments
SharePointdocuments
Slackmessages
Teamsmessages
SalesforceCRMDealCloudCRM
AttioCRM
AffinityCRM
Snowflakewarehouse
Tegusexpert callsSalesforce·AccountNorthwind Cap. Partners
DealCloud·CompanyNORTHWIND CAPITAL (dup #2)Outlook·Fromanna@northwind-capital.com
Granola·NoteH. Capital (Berlin office?)One record · example
SalesforceDealCloudOutlookGranola+5Question
- 1IC memo · Nov 2021 · p. 3
- 2Email · adviser teaser · this week
- 3Email thread · Tom to CFO office · 2019
Every company, person and deal becomes one record.
The same company is spelled nine ways across nine systems. Seisa resolves them into one canonical record and links it to every person, document, deal and decision that touched it: a knowledge graph of how everything in the firm relates.
One record · example
SalesforceDealCloudOutlookGranola+5Question
- 1IC memo · Nov 2021 · p. 3
- 2Email · adviser teaser · this week
- 3Email thread · Tom to CFO office · 2019
Plain-language questions. Cited answers.
Ask what the firm knows and get the answer in seconds, with every claim linked to the source document and page. Multi-hop: from the company to the people who met them to what IC decided and why.
One record · example
SalesforceDealCloudOutlookGranola+5Question
What do we know about Northwind Capital ahead of Thursday?
- 1IC memo · Nov 2021 · p. 3
- 2Email · adviser teaser · this week
- 3Email thread · Tom to CFO office · 2019
Ask another
Infrastructure, not another app
It sits under the tools you already use.
Seisa is the layer between your data and your agents. It does not replace Hebbia, Dust, a data lake or Claude. Those are answer tools, and they have no memory of your firm. Seisa is the memory they should run on, exposed as an MCP server so any agent surface can query the firm’s record directly.
Your data
Stays in the systems that hold it.
OutlookDriveSalesforceAttioDealCloudSlackSnowflakeGranolaData roomsCall notesRead-only, no migration
Every claim links back to the source
One record per company, person and deal.
The memory
Agents we build on it
Surfaces it exposes
Answers with sources
Queries through MCP
Your team
Where the answer lands.
The agents they already use
Claude
ChatGPT
Copilot
GeminiNothing moves and nothing is re-entered. Seisa reads where your data already lives, and the answer lands back in the tools your team already has open.
Why Seisa
Most funds have tried. Most builds died on the same three problems.
Those three problems are the product. We spent the first year on them before building anything a partner would see.
Entity resolution at fund scale
Fifteen years of CRMs, acquired systems and inboxes spell the same company a dozen ways. Resolve it wrong and every answer downstream is wrong. Seisa resolves conservatively and keeps every merge reversible.
A private-markets ontology
Generic graphs don’t know a fund from a deal from a portfolio company, or an IC decision from a meeting note. Seisa’s graph is typed on how PE actually works, so multi-hop questions resolve instead of drifting.
Provenance, not confidence
Nobody trusts a system whose accuracy is a feeling. Every claim carries the document and page it came from, and the whole thing is scored against a golden question set your team writes and owns.
entities in a single firm’s graph
relationships between them
to cross from a company, to who met them, to what IC decided
Scale of a single firm’s graph today. The latency is measured; the two counts are round markers while the graph is still growing.
Why now
Every fund can buy the same model. Nobody can buy your history.
Foundation models are a commodity. The moat has moved to the relationships and the proprietary knowledge a firm already owns. The firms that structure their own history first compound on it; the others start every deal from zero.
A model that has read the whole internet has still never read your 2019 pass memo.
General intelligence is bought. Particular knowledge is accumulated, and only by you.
Forty passes for every purchase. The passes are the dataset, and they are the part nobody keeps.
The reasoning behind a decision not to proceed is the most reusable thing a firm produces and the least often filed.
The fund across the street can license the same model tomorrow. It cannot license fifteen years of your judgment.
Which is why the advantage is not the model. It is whether your own record is in the room when the next deal starts.
The ask
Spend twenty minutes on your own deals.
Bring two or three people from the deal team and one live deal. We connect it beforehand, so the session is the product working on your own documents rather than a walkthrough.