OnPisteOnPiste

Quick Start Guide

Get started with OnPiste in 5 minutes

Quick Start Guide

Get up and running with OnPiste in just a few minutes.

Prerequisites

  • Chrome or Edge browser
  • OnPiste extension installed (Installation Guide)
  • At least one LLM API key

Step 1: Configure Your LLM

  1. Click the OnPiste icon in your toolbar

  2. Click the Settings icon (āš™ļø) in the top right

  3. Navigate to LLM Settings

  4. Add your API key for your preferred provider:

  5. Select models for each agent:

    • Navigator: Handles web interactions (recommend Claude Haiku 3.5 or Gemini Flash)
    • Planner: Creates strategies (recommend Claude Sonnet 4 or GPT-4o)

  6. Click Save

Step 2: Try Your First Task

Let's start with a simple task. Open the OnPiste side panel and type:

Go to news.ycombinator.com and get the top 5 stories

What happens:

  1. The Planner creates a strategy
  2. The Navigator opens the website
  3. Agents extract the information
  4. Results are displayed in the chat

Step 3: Try More Complex Tasks

Research Task

Search for "best wireless headphones 2025" on Google and summarize the top 3 results

Product Comparison

Go to Amazon and find portable Bluetooth speakers under $50 with at least 10 hours battery life

Data Input

Enter "best laptops 2025" in the search box and click search

Understanding the Interface

Side Panel Components

Chat Area

  • Type your commands here
  • View agent responses and progress
  • Ask follow-up questions

Status Indicators

  • šŸŸ¢ Active: Agents are working
  • šŸŸ” Waiting: Processing or awaiting input
  • šŸ”“ Error: Something went wrong

Agent Activity

  • See which agent is currently active
  • View real-time progress updates
  • Monitor decision-making process

Settings Panel

LLM Configuration

  • Add/edit API keys
  • Select models for each agent
  • Test connectivity

General Settings

  • Adjust automation speed
  • Enable/disable features
  • Manage conversation history

Best Practices

1. Be Specific

āŒ "Find a speaker" āœ… "Find portable Bluetooth speakers on Amazon under $50 with at least 10-hour battery life"

2. Break Down Complex Tasks

Instead of one mega-task, use follow-up questions:

  1. "Go to Product Hunt and find today's top product"
  2. "Tell me more about its features"
  3. "Visit their website and get pricing information"

3. Use Natural Language

Write as if you're talking to a human assistant. OnPiste understands context and nuance.

4. Check Agent Progress

Watch the status updates to understand what the agents are doing and when they might need guidance.

Common Tasks

Web Research

Research [topic] and provide key points from the top 3 sources

Price Monitoring

Check the price of [product] on [website] and let me know if it's under [amount]

Data Extraction

Go to [website] and extract all email addresses from the contact page

Text Input Automation

Enter [text] in the search field and submit

Optimizing Performance

Model Selection

Best Performance

  • Planner: Claude Sonnet 4
  • Navigator: Claude Haiku 3.5

Cost-Effective

  • Planner: GPT-4o or Claude Haiku
  • Navigator: Gemini 2.5 Flash or GPT-4o-mini

Zero Cost (Local)

  • Both: Qwen3-30B or Falcon3 10B via Ollama

Tips for Local Models

When using Ollama or local models:

  • Use more specific, cleaner prompts
  • Avoid high-level ambiguous commands
  • Break tasks into clear, detailed steps
  • Provide explicit context and constraints

Troubleshooting

Task Failed

  • Check if API keys are valid
  • Try a simpler version of the task
  • Check console for error messages

Slow Performance

  • Use faster models for the Navigator
  • Break complex tasks into smaller steps
  • Check your internet connection

Wrong Results

  • Be more specific in your prompt
  • Add constraints and context
  • Use follow-up questions to refine

Next Steps

Need Help?

  • Discord: Get help from the community
  • GitHub: Report bugs or request features
  • FAQ: Check common questions and answers