Give Portfolio Managers Investment Memory That Tracks Every Thesis
A PM's mental model of each holding compounds over years. AI tools see it for a single session. MemoryLake gives portfolio managers a structured investment memory layer — theses, analyst notes, filings, conviction levels — so AI assistance carries the full thesis history into every research turn.
Give Portfolio Managers Investment Memory That Tracks Every Thesis
Get Started FreeFree forever · No credit card required
The problem: AI tools see today's question, not the years of context behind it
You asked the AI about a 10-Q surprise. It analyzes the document — but doesn't know you've held the name three years, what the original thesis was, or which earlier red flags you watched. Without memory, AI research output is a shallow read on each new prompt.
How MemoryLake solves investment memory for portfolio managers
Per-name memory
Each ticker has its own memory namespace: thesis, holdings history, analyst notes, conviction changes.
Built-in SEC EDGAR + financial data
3M+ SEC filings, real-time stock/crypto/FX, World Bank, FRED — instantly queryable.
Reflection memory for thesis evolution
How your conviction has changed, what triggered each change, what would force a reversal.
Event memory for catalyst tracking
Earnings, management changes, regulatory events — chronological per name.
Free forever · No credit card required
How it works for investment memory
- Connect — Ingest analyst notes, IC memos, filings, and prior call transcripts.
- Structure — Each becomes typed memory: thesis (background), filings (facts), earnings calls (events).
- Reuse — Before any AI-assisted research turn, the relevant per-name memory loads automatically.
Before vs. after: investment memory
| Without MemoryLake | With MemoryLake | |
|---|---|---|
| Researching a 10-Q surprise | AI reads only today's filing | AI loads thesis + history |
| Analyst rotation | Lose institutional memory | Memory stays with the name |
| Thesis drift over a year | No record | Reflection memory tracks it |
| Audit trail for an IC decision | Manual reconstruction | Versioned memory at decision time |
Who this is for
Hedge funds, family offices, and asset managers using AI in research workflows — where conviction depends on years of accumulated context per holding and the cost of losing that context to AI amnesia is real performance.
Try this use case on your own account
A seven-stock book with one project per holding. A note filed under one holding says its largest customer is bringing its warehouse technology in-house and leaving its leased hub. The weekly review, holding by holding, flags one holding of seven. The same premise check over the whole book flags three: two more theses rest on the same customer, and their own memory never heard the news. One of them was reaffirmed at high conviction a week after the note was already in memory.
What you will see
- Each holding is its own project; every record (thesis, premises, earnings and news notes, conviction updates) is one fact, word for word, with
fact add.fact update --metadatarecords the company each premise depends on and what would break it, and the companies and events each note reports - The weekly review searches each holding's project (
search --projects <holding>) and checks its premises against that holding's memory: 1 of 7 flagged. Nothing is wrong in what it reads; the bad news was simply filed elsewhere - The same check over every holding's
fact listflags 3 of 7 (32% of the book): two premises broke on a note filed under another holding. Run as of the day before the news it flags nothing, but already shows one customer carrying premises of three holdings - The write-back files a cross-reference into each blind holding (
xref_ofin its metadata) and marks the theses; the holding-by-holding review then flags 3, and the holding's own search returns the cross-reference first. Plus a local web UI
You need: a MemoryLake API key · the memorylake CLI · Python 3.9+
Run it
git clone https://github.com/memorylake-ai/memorylake-usecases.git cd memorylake-usecases/investment-thesis-memory export MEMORYLAKE_API_KEY=sk-… python3 demo.py # or: python3 web/server.py
Watch it run · real, unedited recordings
Related use cases
Frequently asked questions
Which financial data is built in?
Which financial data is built in?
SEC EDGAR (3M+ filings), real-time stock and crypto pricing, FX, World Bank, FRED economic data — no additional setup required.
Can a PM team share memory while keeping individual notes private?
Can a PM team share memory while keeping individual notes private?
Yes. Per-name memory at the firm scope; personal annotations stay in private layers with controlled sharing.
Is the memory exportable for compliance review?
Is the memory exportable for compliance review?
Yes. Full export with versioned audit trail per memory item.