Eric's financial copilot
My research agent for the stock market, with three more agents around it. Every call is checked against what actually happened, so you always know what happened and why, and what it learns changes how it decides next time. Built with AI assistance and no programming background.
01 · WHY THIS ONE
I built it because I wanted to know what happened and why: where every number came from, why each call was made, and whether it was right. Four differences from the AI you've already tried, and a receipt for each.
Most AIanswers when you ask.
This oneruns 42 scheduled jobs, about 170 runs a week, and messages me when something matters. No dashboard to check.
Weekdays, 6:00 a.m. Pacific: the morning briefing runs before the market opens.
Most AIsounds just as sure when it's guessing.
This onestamps every number with its source and the time it was fetched, and a missing number stays missing instead of becoming a guess.
A real miss became a rule: an empty filing lookup has to say whether the company didn't disclose it or the lookup failed.
Most AIalways has a pick.
This oneholds every idea to a written bar and acts only when one clears it. When nothing does, you see exactly why.
Sep 27: 6,774 names screened, 5 full workups, best score 7.25 against a bar of 8. It waited.
Most AInever finds out.
This onechecks every call, mine included, against what actually happened and against plain rules with no AI in them. What it learns changes how it decides.
Oct 11: in 8 of 8 hunts, the names one stage turned down beat the ones it kept over the next 20 sessions. So that stage lost its veto.
It's built around one person's rules: mine, written down and read before every run. Built for you, it would run on yours. Tell me what it should do →
02 · HOW IT'S BUILT
The numbers are the order it was built: one agent on a server first, then what it needs to think (3 to 5), the team (6), the schedule that runs it (7), and the parts that keep it honest (8 and 9). Point at or tap a piece, and the diagram shows the system up to that point.
A small rented server in a data center. Always powered, always online. The same kind of machine people use to host a blog.
Not a chatbot, and not always listening. A scheduled program that starts on time, reads its context, makes decisions with an AI model, and writes the results to files.
Every session starts by reading a long instruction file: who it is, its rules, and my own investment reasoning, written down. Replace the file and you have a different agent.
Agents don't remember yesterday on their own. Every run writes its work to shared files: open theses, decisions, the last briefing. The next run reads them first.
Before the market opens, plain scripts pull insider filings, SEC filings, the macro calendar, institutional holdings and congressional trades. Each writes a file the agent reads when it wakes up.
I talk to one of them: a chief of staff on Telegram. It hands deep work to the research agent. A marketing agent reads the news every morning and flags what matters, and a demo agent is built to talk to the public. When something matters, I get a message.
42 scheduled jobs, about 170 runs a week, from the 4:30 data feeds to the Sunday hunt. Predictable, auditable and cheap. Nobody has to remember to run them.
Every call is logged and graded from the next open, against plain rules with no AI in them. A scoreboard rebuilt every weekday decides what the system is allowed to claim.
Every change ships as a package. I read the before and after and say yes; the installer backs up each file, proves the change landed, logs it, and can undo it with one command.
The full system diagram, a weekday's schedule and how the scorekeeping works are on their own page: How it runs →
03 · A DAY
42 scheduled jobs, about 170 runs a week, and nobody has to remember to run them. These are the ones that shape a weekday. Tap any point on the dial to jump to it.
06:00 PT · WEEKDAYS
The research agent reviews the state files and flags what needs a human. It executes nothing on its own.
04 · THE HUNT
My research agent for the stock market, and the one doing most of the work. Six stages, each one cutting the list, and when nothing clears the bar it says no trade and shows why. Two real runs:
That was one run. The full copilot page walks through each stage, the critic's standing questions, and the log of every idea it rejected.
See how it hunts →Want one
I read every one. Response within 48 hours.
Just want to follow the research? Join the copilot waitlist →
Book a Discovery Call