FlowPilot

Inspiration

Repetitive browser tasks are everywhere: filling forms, collecting information, checking websites, moving data between tools, and following the same multi-step process again and again.

We wanted to make automation accessible without requiring users to learn programming, complicated workflow builders, or browser automation scripts.

The idea behind FlowPilot is simple: show the task once, turn it into a workflow, and let AI handle it autonomously.

What it does

FlowPilot turns browser-based tasks into reusable AI-powered workflows.

Users can describe a goal or demonstrate a browser task, and FlowPilot helps turn that process into an executable workflow.

Once created, a workflow can:

• Navigate websites and interact with web pages
• Fill forms and perform repetitive actions
• Work across multiple browser tabs
• Extract and process information
• Run scheduled tasks
• Use AI to make decisions during execution
• Pause for human approval when an important action requires confirmation
• Track workflow execution and results

Instead of repeatedly performing the same task manually, users can create it once and let FlowPilot handle future executions.

How we built it

FlowPilot is built on top of OpenSidekick, an open-source browser-agent foundation.

We used its existing browser automation and agent capabilities as the technical foundation, then focused our hackathon work on turning those capabilities into a more focused autonomous-workflow product.

Our architecture combines:

• AI agents for planning and decision-making
• Browser automation for interacting with websites
• Workflow recording and execution
• Scheduling for recurring tasks
• Human-in-the-loop approval
• Execution tracking and verification
• A user-facing workflow experience

We also use existing open-source libraries and AI/developer tools where appropriate and disclose the upstream technologies used in the project.

Challenges we ran into

One of the biggest challenges was making browser automation reliable.

Websites change, pages load asynchronously, elements move, and the same task can require different decisions depending on what the agent encounters.

We also had to think carefully about autonomy versus user control. An agent that can perform actions automatically should not blindly execute consequential actions.

Another challenge was turning individual browser actions into workflows that can be reused instead of simply replaying a fixed sequence of clicks.

Accomplishments that we're proud of

We built a working concept around a simple idea:

Show it once → automate it repeatedly.

We are particularly proud of combining browser automation with reusable workflows, scheduling, AI decision-making, and human approval into a single product experience.

Rather than creating another generic chatbot, FlowPilot focuses on actually getting tasks done.

What we learned

We learned that autonomous agents are not only about making an AI model smarter.

Reliable automation requires several layers working together:

Planning → Action → Observation → Verification → Human approval

We also learned the importance of building around a real user workflow instead of adding AI simply for the sake of using AI.

Finally, building on open-source infrastructure taught us how to extend an existing system responsibly while clearly distinguishing upstream functionality from our own contributions.

What's next for FlowPilot

We want FlowPilot to become a general-purpose personal workflow engine.

Future versions could include:

• More browser and application integrations
• Better workflow generation from demonstrations
• Automatic error recovery
• More robust verification
• Workflow sharing and templates
• Team workflows
• Secure credential management
• More granular approval controls
• Analytics and execution history
• A marketplace for reusable workflows

Our long-term vision is:

If a person can demonstrate a repetitive digital task, FlowPilot should be able to learn it, understand it, and eventually automate it.

Built With

• agentic
• agents
• ai
• automation
• browser
• development
• github
• javascript
• llm
• mcp
• node.js
• open
• playwright
• react
• source
• typescript
• workflow