Ollie Labs
ProjectDeployed

Digital Employee

An AI agent that learns the context of your organization — its tools, workflows, decisions, and culture — and operates autonomously on its behalf across the web.

Digital AI AgentsBrowser AutomationOrganizational MemoryContinual Learning
The Problem

AI tools are stateless. Every session starts from scratch. Human employees aren't built this way.

A new hire on their first day knows nothing about your organization. By month three, they know which tools you use, which decisions have already been made, who to ask for what, and how things get done. By year two, they carry institutional knowledge that would take months to transfer to someone new.

AI assistants today are perpetual first-days. Every conversation starts with the same blank slate. They cannot accumulate context, carry history, or develop expertise in your specific environment. They are powerful but institutionally amnesiac.

Digital Employee is our attempt to fix this. An agent that learns your organization the same way a human employee does — through repeated interaction, accumulated context, and explicit memory of what has been decided, tried, and learned.

What It Can Do
Browser Automation

Operates the real web

27 production skills across X, LinkedIn, and the broader web. Not mock environments — the same DOM-drift, session-expiry, modal-interruption reality that breaks most automation.

Organizational Memory

Carries institutional knowledge

10,000+ typed memory objects indexed across sessions. Decisions, lessons, events, commitments — all stored with source provenance and retrieved in under 50ms.

Multi-Surface Operation

Works where you work

Operates across Discord, Telegram, and the browser. The same agent that executes research tasks in the browser responds to questions in Slack with the same context.

Self-Healing Skills

Doesn't break when the web changes

An 8-level selector reliability hierarchy that recovers from CSS obfuscation, DOM reshuffles, and React re-renders. When a selector breaks, the system heals itself.

The Selector Ladder

Eight levels of selector reliability. The agent tries each in order until one works.

The most common way browser automation breaks is a brittle selector. CSS classes get obfuscated. DOM structure changes. React re-renders move elements.

Our solution is a reliability hierarchy. Every selector attempt starts at the highest-reliability level (ARIA) and falls back through the ladder until something works. When a lower-level selector succeeds, the system logs the healing event and updates its cache.

The result: 27 production skills that have survived months of DOM changes, site redesigns, and A/B test shuffles without manual intervention.

L1ARIA role + labelHighest
L2Semantic HTML (button, input)Very high
L3Stable data-testid attributeHigh
L4Text content matchGood
L5Structural positionModerate
L6CSS class selectorLow
L7XPathVery low
L8Coordinate click (CDP)Last resort
Running in Production
27
production automation skills
10K+
memory objects indexed
<50ms
memory recall latency
300+
sessions with continuity
Technology
Automation
Playwright + CDP
not Puppeteer
Memory
SQLite + Xenova
not Postgres / not OpenAI
Intelligence
Claude Haiku
for extraction + reasoning
Surfaces
Discord + Telegram
where you already work

Every choice is a position.

Want an agent that knows your organization?

We are most interested in organizations with repeatable browser-based workflows that currently require a human to remember the context of what has been done before. That is the problem Digital Employee is built to solve.