Getting Started
The fastest path from zero to a working AI research process. No custom setup required — just a Claude subscription and these five steps.
Pick your model
Start with Claude. It handles long context well — which matters when you're feeding in fund docs, research notes, and working context all at once. The free tier works to start, but Claude Pro is worth it if you're using this seriously.
You don't need a custom setup yet. A browser tab is fine. The system you're about to build works with nothing more than a Claude subscription.
Don't overthink the model choice. The system matters more than which AI you use. You can swap models later — get the structure right first.
Create your three context files
AI has no memory between sessions. Every time you open a new chat, it knows nothing about you, your portfolio, or what you were working on yesterday. The fix is simple: three markdown files you maintain yourself.
Create a folder called investment-os/ somewhere you can find it easily. Inside it, create these three files:
investment-os/ ├── working-context.md # what you're working on right now ├── hot-context.md # what happened in the last 48h └── topics/ # one file per fund or subject
These files are your AI's memory. Every session starts with pasting them in. Every session ends with updating them. That's the whole system.
Write your first working-context.md
Keep it short. One active task, where you are, what's next. This is your live scratchpad — not a journal, not a report.
# Working Context Last updated: [date] ## What I'm working on Reviewing Fasanara Q1 factsheet — checking yield vs prior quarter ## Where I am Downloaded PDF. Extracted key numbers. Need to compare to hurdle rate. ## What's next - [ ] Update returns DB with Q1 figures - [ ] Check if yield still clears the bar at current gilt rates ## Open questions - Has NII coverage ratio improved or declined?
Start every session the same way
Open Claude. Paste this at the top of every new chat:
Here's my current context. Please read it and we'll continue from here. [paste working-context.md] [paste hot-context.md] [paste relevant topic file if working on something specific]
Two minutes of context pasting saves twenty minutes of re-briefing. Do it every time without exception.
End every session with an update
Before you close the tab, run this prompt. It takes 30 seconds and means you never start from scratch again.
Before we finish, please update my context files: 1. working-context.md — where we are now, what's next 2. hot-context.md — 2-3 bullet summary of what happened today Output each file in full so I can copy it.
This is the manual version of the night shift. At the full stack tier this runs automatically. Here it's five minutes at the end of each session. Same result.
When you've outgrown this
Once you're using the context system daily and hitting its limits — spending time on manual data lookups, wishing the AI could watch something for you, wanting to query ingested docs — that's when the next layer becomes worth building.
The knowledge wiki is where most people go next. Ingest your fund docs, quarterly letters, and research notes. Make them queryable. That's when it stops feeling like a better search engine and starts feeling like a team member.