Automated Investment Research and Portfolio Tracking

The weekly research grind

You know the routine. Sunday evening or Monday morning, you sit down with a browser full of tabs. Morningstar for fund performance. Yahoo Finance for price history. Your brokerage dashboard for holdings. Maybe a Google Sheet where you track things manually. You are cross-referencing numbers, comparing YTD returns against last quarter’s fact sheets, eyeballing risk metrics, and trying to spot trends across your portfolio.

This takes two to three hours. Every single week. And if you skip a week — because life happens — you lose the thread. Trends disappear into gaps. The comparison you were tracking between SPY and VTI becomes meaningless when you have missing data points. Inconsistency is the enemy of good portfolio management, and manual processes are inherently inconsistent.

The irony is that individual investors have access to more data than ever. The problem was never the data. It was always the labor of collecting, structuring, and comparing it on a regular cadence.

What automated portfolio tracking looks like in practice

One LikeClaw user set up a weekly Sunday analysis tracking SPY, QQQ, and VTI. The agent reads their Q3 2025 fact sheets from the workspace as a baseline, searches for current performance data, and produces a structured report with YTD returns, 1-year and 3-year annualized returns, expense ratios, top holdings changes, risk indicators, and Sharpe ratios. After multiple weeks of consistent execution, they have a complete performance history saved to their workspace — without opening a single financial site manually.

The reports are not summaries scraped from a single source. The agent searches the web for current data across multiple financial sites, cross-references the numbers, and compares everything against the baseline files already in the workspace. Week-over-week changes are calculated and flagged. If an expense ratio shifts or a top holding drops out of a fund, the report notes it.

This is not a static dashboard. It is an active research agent that does the work you used to do by hand.

Iterating until the report is right

The same user refined their scheduled task multiple times — adjusting analysis requirements, adding new metrics like sector allocation breakdowns, tweaking the report format to highlight risk-adjusted returns more prominently. Each iteration improved the output because the agent operates on persistent workspace files. Your baseline data, your previous reports, your configuration preferences — all of it persists across sessions.

This is a workflow that gets better over time. You are not starting from scratch every week. You are building on a growing dataset that makes each new report more useful than the last. The workspace is your institutional memory.

Ad-hoc research when you need it

Scheduled reports handle the routine. But sometimes you need answers now. A stock you hold just reported earnings. A sector is rotating. Someone mentions an ETF you have not looked at before.

LikeClaw’s deep-research mode handles these one-off questions. Ask “What is the current outlook for SOFI given my small position and tolerance for downside risk?” and the agent searches for recent analyst coverage, earnings data, price targets, and risk factors — then synthesizes a research brief tailored to your stated context. No need to configure a new schedule. Just ask the question and get a sourced, structured answer.

This is the same capability powering the scheduled runs, used on demand. Same web search. Same sandbox execution. Same workspace for storing results.

Python analysis in a secure sandbox

Some analysis goes beyond what web search can deliver. You want a custom Sharpe ratio calculation across your portfolio. A correlation matrix between your holdings. A Monte Carlo simulation for retirement projections. Visualization of sector exposure over the last six quarters.

The E2B sandbox runs Python with pandas, numpy, matplotlib, and other common financial libraries pre-installed. The agent writes the analysis code, executes it in an isolated container, and returns charts and calculations to your workspace. The container is created for your task and destroyed after — your portfolio data never persists outside your encrypted workspace.

If you are already using LikeClaw for data analysis, investment research is the same underlying capability applied to financial data. Same sandboxed environment, same persistent workspace, same multi-model access.

One account vs. the subscription pile

The average professional spends $133 per month across AI subscriptions, and research from Arsturn found that 42% of that spending is wasted. If you are paying for a market data terminal, a charting tool, a portfolio tracker, and a separate AI assistant for research — that adds up fast.

LikeClaw consolidates the research and analysis layer. Credits-based pricing — cheaper models cost fewer credits, premium models cost more. You get 20,000 free credits at signup and 5 free generations per day. Buy credit packs as you need them. No subscriptions, no commitments, no runaway costs. Compare that to open AI agent frameworks where users report spending $50-750/month with no cost controls.

For a use case like weekly portfolio analysis, transparent pricing is not a convenience. It is a requirement. You need to know the cost of running this every Sunday for the next 52 weeks, not discover it retroactively.

Building a research system, not running a task

The real value is not any single report. It is the system you build over time. Week one, you have a baseline and a first analysis. Week ten, you have a trend line. Week twenty-six, you have six months of structured, comparable data points — all generated automatically, all stored in your workspace, all searchable and referenceable.

Combine this with task automation and you can chain investment research into broader workflows: run the weekly analysis, compare against your target allocation, flag rebalancing opportunities, and push a summary to your email. Each piece runs in its own sandbox. Each piece is scheduled independently. The system runs whether you are at your desk or not.

That is the difference between using AI as a search engine and using AI as a research analyst. One answers questions when you ask. The other does the work on your schedule, builds on what it learned last week, and delivers results you can act on.