Browser agents that work with legacy portals (not against them)

Legacy enterprise portals are where clean API dreams go to die. Multi-factor auth, CAPTCHA screens, dynamic form IDs that change on every page load, and JavaScript that breaks if you breathe on it wrong.

But this is where 80% of real business data lives. Client portals, vendor dashboards, compliance systems built in 2003 and "upgraded" with jQuery in 2015.

We spent months trying to scrape these systems with traditional tools. Selenium scripts that broke every week. Custom scrapers that required a PhD in DOM archaeology. Then we found the stack that actually works for gated portal automation.

The three-layer approach:

• Skyvern as the browser agent — it navigates like a human, not a script
• Screenpipe for persistent context memory across sessions
• OpenBB for verified financial data to cross-check what we extract

Here's what makes this different: Skyvern doesn't rely on CSS selectors or XPath. It uses computer vision to find form fields, buttons, and navigation elements. When the client portal redesigns their layout (again), your automation doesn't break.

Screenpipe runs locally and remembers everything your agent saw on screen. Previous error messages, successful login flows, document locations. No more "I don't remember how we accessed that report last week."

OpenBB provides the reality check layer. When your agent extracts financial data from a gated portal, you can verify it against public market data, SEC filings, or other authoritative sources.

Example workflow:

1. Skyvern logs into client portal
2. Navigates to quarterly reports section
3. Downloads latest P&L statement
4. Screenpipe recalls similar document locations from memory
5. OpenBB cross-checks revenue figures against public filings
6. Flags discrepancies for human review
7. Generates summary with confidence scores

Critical safety rules: Never allow browser agents to handle payments, wire transfers, or document approvals without explicit human confirmation. Configure Screenpipe to exclude sensitive windows (banking, password managers). Always cross-check extracted financial data against primary sources.

The key insight: don't try to make legacy portals modern. Build agents that work with legacy portals as they are. Skyvern handles the brittle UI layer, Screenpipe provides institutional memory, and OpenBB keeps your data honest.

This isn't about replacing human judgment — it's about automating the tedious portal navigation and data extraction that burns 6 hours of analyst time per week.

We're seeing teams use this stack for insurance claim portals, vendor payment systems, regulatory filing platforms, and client reporting dashboards. Anywhere APIs don't exist and screen scraping breaks monthly.