The thesis drifted
News, a competitor, or plain execution moved. Now you defend the position for a different reason than the one you bought it for.
Open-source · Self-hosted alpha
Easy to buy.
Hard to know when to sell.
Plenty of tools help you find the next 10x stock. This one is for the harder question: what to do with the positions you already own.
Why I built it
You bought it for a reason.
Does that reason still hold?
Once you own something it stops being an idea. It has a price you paid, a chart you have watched, and a story you have defended out loud. PortVise runs the argument again on every holding, without the sunk cost.
The winner became the risk
Your best holding keeps winning until it is a quarter of the book, and the whole portfolio now rides on one story.
A drawdown feels like being wrong
So you sell into it, even when nothing about the case has actually changed. Volatility is not the same thing as a broken thesis.
How a review runs
Nine agents. One decision you can audit.
A single prompt gives you a polished answer. It also blends facts, calculations, and opinion until you cannot tell where the conclusion came from. PortVise splits the review into three passes and keeps the reasoning channels apart, so they get the chance to disagree.
Collect
Prices, headlines, your portfolio, and the macro backdrop.
Challenge
Separate passes for risk, regime, and theme, then a check for contradictions.
Decide
Findings become ranked actions: reduce, exit, hedge, rotate, add, or monitor.
Planner
Sets the review plan and search priorities.
Data
Fetches live prices, headlines, and macro data.
News
Builds the macro and market context.
Risk
Concentration, factor exposure, and what breaks in a bad scenario.
Regime
Finds past periods that rhyme with this one.
Theme
Asks whether several holdings are saying the same thing.
Validation
Flags contradictions and missing inputs before anything ships.
Allocation
Argues the opportunity cost of sitting in cash.
Manager
Weighs it all and writes the ranked actions.
Decision memo
Make every position earn its place.
The memo says what to do, what the main risk is, when to look again, and what would change the answer. If it cannot make that case, it says so.
- Reduce, exit, hedge, rotate, add, or monitor
- A reason and a size for every position
- The case for doing nothing, written down
Verdict
This week
Keep the thesis. Reduce the concentration.
The AI case still holds up. The problem is that these holdings move together, so the downside is larger than any single thesis suggests.
- Posture
- Selective trim
- Primary risk
- Factor concentration
- Revisit
- After earnings
UrgentReduceNVDATrim toward 18-20%
WatchHoldMSFTThesis remains intact
Ten tickers can still be one bet.
Owning a lot of names does little when they all ride the same rates, the same liquidity, the same theme. This measures that.
History rhymes often enough.
Match today's macro state against past periods and see what the portfolio did next. Context, not a forecast.
Match78%Q4 2018
Avg return+4.8%21 days
Max drawdown−9.2%21 days
The PortVise Daily Brief
Three small portfolios.
One review after the close.
Three portfolios in different styles, held in the open. PortVise reviews all three after the U.S. close and sends what it would do about them.
PortVise
Daily Brief
Monday, June 29, 2026
Today's Brief
The verdict and the ranked actions for every PortVise portfolio, sent after the U.S. close.
Portfolio 02
AI Infrastructure
Reduce concentration · urgent · Single-name dominanceNVDA and AMD behave like one 50% position. In a high-volatility regime that is a single-name failure mode wearing two tickers.
reduce · NVDA · cap exposure at 10-12%
The largest single risk driver, and half the portfolio once you count AMD.
add · TLT · add a 10-15% allocation
Duration hedges the portfolio's high beta if rates fall.
Join the distribution
Read the same brief
I read every evening.
One email after the close covering all three portfolios: the verdict, the reasoning behind it, and what to do next. There are no trade alerts and no intraday noise.
Run it your way
Your machine, your models,
your portfolio data.
Point PortVise at any OpenAI-compatible endpoint, whether that is a local LLM or a bigger hosted model. Run the web app with Python, or put the whole thing in Docker.
MIT licensed
OpenAI-compatible
Cloud or local models
CLI and MCP access
portvise / quickstartbash
# Install dependencies
$ uv sync
$ npm install
# Start the portfolio advisor
$ uv run python run.py
✓ PortVise running at http://localhost:7860Review the evidence
I open-sourced this because
I want to hear where it is wrong.
Follow the three portfolios after every market day, or read the code and tell me what it is missing.