Changelog

# 11.3 GPT-5, o3 Models & ChatGPT Teams

AI MODEL UPGRADES

• Integrated GPT-5 and o3 models as default connections for HARPA AI CloudGPT Connection.
• Added o3-mini model support for cost-effective processing.
• Added support for Claude 4.1 models.

CHATGPT TEAMS SUPPORT

• Full support for ChatGPT Teams functionality, enabling collaborative AI usage within organizations.

CONTEXT & FILE HANDLING

• New ChatGPT context button for file uploads with loading indicators.
• Better international text support.

# 11.2.3 Performance & Stability

PERFORMANCE IMPROVEMENTS

• Enhanced data persistence and recovery.
• Better error handling and recovery mechanisms.

# 11.2 GRID API & Statistics System

GRID API ENHANCEMENTS

• New "Run AI Prompt" functionality in GRID API for advanced automation.
• Enhanced validation, error handling, and endpoint optimization.
• Improved controller optimization for better performance.

FEEDBACK SYSTEM

• Enhanced uninstall feedback page for user insights.

UI/UX IMPROVEMENTS

• Fixed YouTube summary URL handling and transcript fetching.
• Improved alert styles and UI element positioning.
• Enhanced dark/light theme support with better theme switching.
• Fixed input responsiveness and explorer button interactions.

# 11.1 Connection & Command Updates

CONNECTION IMPROVEMENTS

• Fixed connection issues and improved stability.
• Better token credit calculations for thinking modes.
• Improved API connection handling.

COMMAND ENHANCEMENTS

• Updated YouTube video summary command.
• Increased command size limit to 20MB.
• Improved command validation and error handling.

# 11.0 Thinking Mode & Perplexity Integration

THINKING MODE

• Added "Think" button for reasoning models.
• Support for thinking/reasoning modes in AI responses.
• Enhanced token calculations for thinking modes.

PERPLEXITY INTEGRATION

• Full Perplexity API integration with Pro and Deep Research models.
• Perplexity search context and citation rendering.
• Links rendering and improved tokenizer for Perplexity.

# 10.9 Advanced Model Support

NEW MODEL INTEGRATIONS

• Groq Llama models support including versatile variants.
• Gemini models support via API.

GLOBAL PARAMETERS

• New global parameters editor in Settings Tab.
• Global parameters tooltips for better understanding.
• Easier parameter management.

# 10.8 YouTube POI & Context Menu

YOUTUBE ENHANCEMENTS

• YouTube Points of Interest (POI) - see video highlights on timeline.
• Improved YouTube transcript fetching.
• Fixed YouTube compatibility issues.

CONTEXT MENU IMPROVEMENTS

• New context menu options and summarize embedded buttons.
• Fixed context menu rendering issues.
• Enhanced context menu for various websites.

# 10.7 PDF Support & File Handling

PDF SUPPORT

• Full PDF file support with drag & drop functionality.
• PDF rendering in chat history and attachments.
• PDF summary command for document processing.
• PDF content insertion into prompts.
• PDF validation with clear error messages.
• Use PDFs as context in your conversations.

FILE TYPE SUPPORT

• Added DOCX to HTML conversion using Mammoth.js.
• XLSX file support for spreadsheet processing.
• Support for text file types (text/plain, text/csv, etc.).
• Better file type recognition.

ATTACHMENTS SYSTEM

• New attachment view with improved UI.
• Better attachment handling during command execution.
• Enhanced file validation and upload process.
• New "Add files" button in chat.
• Attachments stay available across chat sessions.

# 10.6 AI Agent Updates & Models

AI AGENTS OPTIMIZATION

• Renamed and optimized agent models.
• Smart Agent now default over Fast Agent.
• Deprecated Fast Agent in favor of improved models.

MODEL UPDATES

• Claude 3.7 web session connection fix.
• Removed deprecated Sonnet 2.1 and Opus3.
• Optimized model usage for cost efficiency.

# 10.5 Email & Automation Features

EMAIL FEATURES

• Draft email replies support for Outlook.
• Better email handling and formatting.

AUTOMATION IMPROVEMENTS

• Daily autoblogging guide for WordPress.
• WordPress post automation fixes.
• Zapier integration enhancements.
• Fixed GRID panel display issues.

# 10.4 UI Modernization

CHAT UI IMPROVEMENTS

• New chat action bar with better controls.
• Improved attachments rendering in input.
• Updated page attachment view.

# 10.3 Connection Management

CONNECTION FEATURES

• HARPA OpenAI API update.
• HARPA Anthropic API support with thinking modes.
• Connection icons for better visualization.

SPACE CONNECTIONS

• Improved space connection handling.
• Enhanced connection validation.
• Fixed space switching issues.

WEBSITE COMPATIBILITY

• Fixed Google Docs compatibility.
• Fixed Notion.so compatibility.

# 10.2 Performance & Optimization

PERFORMANCE IMPROVEMENTS

• Better handling of PDFs in chat history.
• Improved website compatibility.
• Fixed CloudGPT connection errors.

COMMAND IMPROVEMENTS

• Better PDF processing in commands.
• Improved attachment handling.
• Fixed scheduled AI commands.
• Enhanced command reliability.

# 10.1 Grid Guides & Documentation

GRID DOCUMENTATION

• Multi-language Grid guides (IT/PT/ES).
• Zapier WordPress webhook guide.
• No-cost WordPress blog automation guide.
• Grid REST API reference updates.

INTEGRATION GUIDES

• Zapier integration documentation.
• WordPress integration improvements.

# 10.0 Major Platform Updates

NOTIFICATION SYSTEM

• New in-app notification system.
• Update notifications with better user experience.

PLATFORM IMPROVEMENTS

• Better error handling for CloudGPT.
• Improved extension stability and compatibility.
• Enhanced security and privacy protection.

# 9.9 New Models

MODEL ADDITIONS

• Llama 4 models support.
• Sonnet 4 model integration.

# 9.8 CloudGPT: DeepSeek + Llama

CLOUDGPT ENHANCEMENTS

• DeepSeek R1 and V3 models support, hosted in the United States.
• Llama 3.3 models integration.
• Data privacy ensured - no logging or training on user data.

YOUTUBE IMPROVEMENTS

• Enhanced YouTube transcript API.
• Improved YouTube video summaries.
• Better transcript caching and details fetching.

# 9.7 PDF Support

• HARPA now lets you use PDFs from both the web and your computer's storage.
• You can drag and drop any PDF file from your computer to: use it for context in your prompts, create summaries or ask questions.

# v9.6.2 CLOUDGPT 4o-mini Optimization

CLOUDGPT 4O-MINI FIX

• CloudGPT 4o-mini connection now uses about 3 times fewer AI tokens and fully matches the pricing of other OpenAI Models, allowing you to use your AI tokens longer.

DEEPSEEK API SUPPORT

• DeepSeek V3 is a cost-effective alternative to GPT-4, Claude 3.5 Sonnet and Haiku. You can connect it to HARPA now - see the guide here: Connecting HARPA to DeepSeek API.
• Other DeepSeek models, including R1 with reasoning capabilities, can also be connected following the instructions. Remember to update the chat model name in settings according to DeepSeek's LLM integration instructions.

BUG FIXES

• Bug fixes and improvements, including fixed square bracket rendering, iframe blocking problems, SMTP email sending & more.

# v9.6 GRID API & Connections Sync

HARPA GRID API

• With HARPA GRID API you can connect your Web Browser to a variety of services, such as Google Sheets, Wordpress, GMail, to automate workflows and data processing.
  • GRID complements existing business processes and systems (CRMs, ERPs) and can be integrated in minutes via IFTTT systems (Zapier, Make.com/Integromat, n8n, Microsoft Power Automate, Workato, UiPath) or through its fast, easy-to-use REST API.
  • Learn more: https://harpa.ai/grid/web-automation

CONNECTIONS SYNC

• Your API and CloudGPT connections settings are now stored in the cloud.
• When you activate a new device or browser, or reinstall HARPA AI, your API keys and CloudGPT settings will automatically restore.

TEAM SPACE CONNECTIONS

• CloudGPT and API connections are shared with every space member. Only the space owner can modify them. This makes teamwork easier and helps control AI token usage, system prompts and more.

# v9.5 Automation Library

Easily share your AI-powered web automations with others, either via direct links or publicly in the AI Automations Library.

LIBRARY

• Grab helpful AI commands and Web automations from the Library.
• Publish your commands to the Library.

IMPROVEMENTS

• Support for a new {{thread}} parameter to facilitate Telegram, Discord, WhatsApp, Reddit, YouTube, Facebook and Twitter commands and automations.
• Bugs fixes and improvements.

# v9.4 Side Panel

HARPA can now be rendered in a SidePanel. SidePanel does not obstruct website content and persists HARPA window across navigations.

SIDE PANEL

• Optionally switch HARPA UI to the SidePanel view.

# v9.3 Command Sharing

Commands can now be shared publicly from the Command editor.

COMMAND SHARING

• Share AI commands and prompts with a few clicks.

BUG FIXES

• Bugs fixes and improvements, including fixed LaTeX rendering, autolabel fix, optimized chat storage and more.

# v9.2 Chat Cloud Backup

Chats can now be backed up to secure HARPA cloud and synced across devices, while Vision support has been added to the API keys connection.

CHAT CLOUD BACKUP

• Backup your custom commands to the cloud and synch across devices.

• To enable this feature, remember to turn on the "Backup chats to the cloud" option in HARPA Settings.

• Chats are stored per Account and never shared.

VISION COMMANDS

• 20+ AI commands redesigned to support visual area grabbing.

API KEY CONNECTION VISION SUPPORT

• API Key connections now support AI Vision features. To get started, make sure that your chosen LLM supports AI Vision to avoid errors.

• Both OpenAI API keys and other providers, such as OpenRouter, are supported.

# v9.0 AI Vision

Press Alt+V (^V on Mac) and capture area on webpage: answer questions, find and fix mistakes, explain, summarize, translate, pass exams, finish writing, refine emails and much more.

Keep in mind that Vision features only work with LLMs that support image processing.

• AI VISION

  HARPA supports AI Vision features both over ChatGPT web session and CloudGPT connections.

• IMAGE UPLOAD

  Drag & drop any image from your PC to chat.

• CAPTURE VIEW COMMAND

  • Press Alt+V on Windows or ^V on Mac to capture and send screenshots to AI.
  • Capture View command allows fixing mistakes, solving tests, extracting text, translating, explaining and answering questions in images.

• VIEW PARAMETER

  New {{view}} parameter allows attaching page screenshot in AI commands.

• TONE PARAMETER

  • Declare how AI should mimic your writing to improve quality of generated emails and messages.

  • Tone affects the style of AI Responses in commands like /Reply, /Rewrite, /Finish writing and more.

  • You can change the {{tone}} at any time in settings or by using the /Set tone or /Setup commands.

    

• GPT 4O MINI

  Added support for GPT 4o Mini both in Web Session and CloudGPT connections.

# v8.10 Dynamic Parameters

Use {{dynamic}} parameters in your automations. Commands improved. Integrated Claude 3.5 Sonnet.

• DYNAMIC PARAMETERS

  HARPA supports {{dynamic}} parameter names in automations for complex workflows.

• SEEK TIMESTAMPS IN VIDEOS

  /Ask command now reports YouTube timestamps, to references where the question was answered.

  

• SUMMARY FOLLOW-UPS

  /Summary command now lists follow-up questions, saving time on research.

• CLAUDE 3.5 SONNET

  Claude 3.5 Sonnet beats most AI model benchmarks while being faster. The model is available for Web Session, and Cloud GPT.

# v8.9.1 Command Hotkeys + App Shortcuts

• COMMAND HOTKEYS

  Call HARPA by pressing slash / twice, followed by a command hotkey: S for Search, E for Explain and F for finish writing etc.

  It works on selected text, input boxes, or any webpage you visit.

• APP SHORTCUTS

  

  You can customize these shortcuts in the app settings: Click in the input with the current key combination, then press a new modifier and a key

  Reset the values to default clicking the RESET button.

# v8.9 Function Calling

• FUNCTION CALLING

  • Functions are now available for OpenAI API and Cloud GPT connections, allowing you to instruct AI to make decisions and create complex custom AI agents.

  • Function declarations follow the OpenAI function specifications.

  • In HARPA, you can declare functions in the GPT step. The function can be selected based on the input you provided within that step.

• PAGE SHERLOCK

  The new /Page Sherlock command performs quick SEO analysis, extracts web page metadata, checks for typical SEO problem such as broken links, missing attributes and headings, measures page loading time and resource size. Analysis can be downloaded as JSON and inspected further.

# v8.8 AI Agents

• AI AGENTS

  We have added Fast & Smart AI Agents as premium GPT-based AI connections. Agents can search the web, analyze webpages, and launch commands for you.

  • Fast AI Agent is powered by cost-effective GPT-3.5, and is great for most tasks, handles user requests quickly and affordably.
  • Smart AI Agent is powered by new and smart GPT-4o connection. It's better handles tasks requiring in-depth analysis.

  To try new agent, change your connection to a Fast AI Agent by clicking the connections switch in the bottom left corner. - Ask any question, like what the weather is like in your city today, then HARPA will perform a web search for you. - If reading an article, ask to create a summary and HARPA will run the /summarize command automatically.

  Currently, the summary, research, explain, repurpose, extract and help commands are supported. Additionally, the agent automatically performs web search and webpage analysis when needed.

• GPT-4O CONNECTION

  PT-4o was added to CLOUD GPT, API Keys and Web Session Connection. GPT-4o model is twice as fast and half the cost of GPT-4. Note the GPT-4o web session connection is rate limited by OpenAI.

# v8.7 GPT-4o

• YOUTUBE POI

  /youtube-summary command overlays video timelines with video content summaries.

• GPT-4O CONNECTION

  PT-4o was added to CLOUD GPT, API Keys and Web Session Connection. GPT-4o model is twice as fast and half the cost of GPT-4. Note the GPT-4o web session connection is rate limited by OpenAI.

# v8.6 Grab, Echo, Manifest v3

• MANIFEST V3

  HARPA has been updated to comply with Google's latest requirements, including the implementation of Manifest v3.

  This update ensures that data processing and security standards are up to date.

• ECHO PARAMETERS IN CHAT

  Write a {{parameter}} in chat and press Alt+Enter (or ⌥ +Enter), to get the parameter value.

• GRAB PARAMETER

  Allows you to take value of an HTML element from a web page, whether it's an h1, body, or href, for example: {{grab h1 take=className}}. The take argument specifies data to extract. Read more about parameters.

# v8.5 AI Commands Scheduling

AI Commands can now be scheduled and run on timer in background.

HARPA now shows summaries on top of the YouTube video timelines as you hover over them. Summarization quality has been improved on CloudGPT connection.

• AI COMMAND SCHEDULING

  AI commands can now be scheduled in Automation tab. HARPA will run your custom AI command at specific intervals.

• PREFILL STEP

  New prefill step allows you to prefill the input field with a value before sending a prompt.

• CHAT BOOKMARKING

  Click chat emoji to pin it to the top. Bookmarking a chat keeps it visible and prevents accidental deletion.

• GMAIL AUTOLABEL

  Use /autolabel-gmail command to quickly go through your emails, categorise and assign labels.

• JS STEP TIMEOUT

  RUN JS step now has a timeout option. Set a maximum time to execute long running scripts.

• EXPLAIN & ASK FOLLOW-UPS

  Explain and Ask commands now feature auto-generated quick follow-up questions.

• API CONNECTION HEADERS

  Point HARPA to your local AI model by setting custom headers in API connection.

• CHATGPT CONNECTION FIX

  Free ChatGPT Web Session connection has been re-implemented and restored for the new communication protocol.

# v8.4 Search Agent v2

Updated Search Agent command, new CloudGPT options, bug fixes & more.

• SEARCH AGENT v2

  It is now twice as fast and smarter, efficiently crawling web pages and YouTube transcripts to extract relevant content.

  Added dynamic options for follow-up questions that work in one click.

• CLAUDE AI IN CLOUD GPT

  Switch to Claude AI Haiku, Sonnet or Opus models in your CloudGPT connection.

• STEP CONDITIONS UPDATE

  Now you can add multiple conditions to run a custom command step, for example:

  {{start-menu}} = quick
  {{start-menu}} = detailed
  

# v8.3 Email Assistant

Embedded buttons for Gmail, new Finish Writing command, custom commands testing functionality and more.

• FINISH WRITING COMMAND

  Streamlining email, reply, and comment generation. Draft a simple reply in any web page and expand it to a finished reply.

• GMAIL BUTTONS

  Embedded buttons in Gmail UI: Summarize and Finish Writing

• ISOLATED MODE

  Token usage on Cloud GPT connection has been optimized.

• NEW PARAMETERS

  {{email}} parameter for parsing email content from Gmail, Outlook, Protonmail, and Fastmail. {{input}} parameter for the content of the input box.

• UPDATED PASTE STEP

  HARPA can now simulate typing or replace the text in the input box.

# v8.1 CloudGPT

HARPA v8.1 brings CloudGPT, a fast and reliable Large Language Model connection.

• CLOUD GPT

  Switch to CloudGPT 3.5 or 4 model in the Connections tab. Billed per token usage.

• CLONING API CONNECTIONS

  Set up multiple API KEY connections to quickly switch between different API keys.

• CLAUDE AI 3 MODELS

  Support for new paid and free Claude 3 models: Opus, Sonnet, Haiku.

# v8.0 Premium Features

HARPA v8.0 brings the first set of premium features, HARPA X and HARPA S plans.

• COMMANDS CUSTOMIZATION

  Predefined commands can now be cloned, edited and customized.

• CHAT HISTORY

  Chat history is stored locally. HARPA allows switching and reusing old chats across tabs.

• CLOUD STORAGE

  All custom commands are securely stored in the cloud and accessible on any device under your HARPA AI profile.

• TEAM SPACES

  Create spaces for your team, organization or family and share commands with your space members.

# v7.3 Search Agent & GPTs

HARPA v7.3 brings custom GPTs, page crawler, deep search agent and a set of new commands.

• UPDATED COMMANDS

  Text generation commands now come with embedded search and URL integration options.

• SEARCH AGENT COMMAND

  /search-agent command forms deeply informed search responses, thanks to integrated crawler.

• EXPLAIN COMMAND

  /explain command breaks down complex topics and acronyms, researches leads, and pulls data from the web.

• SENTIMENT ANALYSIS COMMAND

  /sentiment-analysis command categorizes reviews of a product and provides an overall rating.

• GPTs

  ChatGPT+ users now have an access to GPTs directly in HARPA.

• PAGE CRAWLER

  {{page}} parameter now supports query, url and limit options, e.g. {{page Connect HARPA to OpenRouter https://harpa.ai limit=10%}}

# v7.2 Reply tone, TL;DR, OpenRouter

HARPA v7.2 brings support for OpenRouter.AI, new commands and /reply tone customization.

• REPLY TONE

  Teach AI to reply to emails in your own voice by changing the tone used by /reply command.

• OPENROUTER CONNECTOR

  Connect HARPA to 20+ LLM models including LLAMA, MistralAI, Claude 2.1 API and even Perplexity via OpenRouter.ai.

• NEW COMMANDS

  Added /tldr, improved /summary, /rewrite and other commands.

# v7.1 DALL·E, ClaudeAI v2.1, Export

HARPA v7.1 brings ChatGPT+ DALL·E, ClaudeAI v2.1, chat EXPORT step, LSI keyword extraction, bug fixes and new commands.

• EXPORT STEP

  HARPA can now export extracted data to files with the new EXPORT step.

• LSI KEYWORD EXTRACTION

  Added automatic SERP LSI keyword enrichment to /article-creator, /outrank-seo-article, /blog-post-writer, /repurpose-text, and /repurpose-video.

• NEW COMMANDS

  Added /youtube-channel-analysis, /extract-youtube-transcripts and /language-wizard for language study.

• DALL·E & CLAUDE AI 2.1

  Integrated ChatGPT+ DALL·E for image rendering. Added support for Claude AI 2.1 model.

• CLAUDE AI CONNECTION

  Resolved connectivity issues with Claude AI.

# v7.0 Web Automation

HARPA v7.0 brings over 10 new features for creating AI-powered web automation workflows: new steps, conditions, loops, jumps, choice options and more.

• ASK STEP OPTIONS

  You can now declare choice options in your AI commands and build complex multi-step dialogues.

• SERP PARAMETER

  Use new {{serp query}} parameter in chat to search web.

• CALC STEP

  • CALC step brings 15+ functions for setting parameters, extracting prices and code blocks from GPT responses.

  • Utilize regular expressions and work with lists in your custom AI commands.

• GROUPS & LOOPS

  • Declare step groups and loops to iterate over list variables.

  • Using loops you can now build mass-automation workflows with HARPA AI.

• WAIT STEP

  Instruct HARPA to wait for a certain amount of time, text or element to appear or JS function to resolve, before proceeding to the next step.

• CONDITIONS, LABELS & JUMPS

  Control command flow execution with step conditions and jumps.

• REQUESTS INTERPOLATION

  REQUEST step URL, Body and Headers are now properly interpolated with parameters. Send data to web APIs such as Zapier, Make.com and n8n, or fetch web pages.

• COMMANDS CLONING

  Clone your custom AI commands with a click of a button.

• PARAMETERS UPDATE

  Parameter names updated for brevity: {{selection}}, {{url}}, {{domain}}, {{title}}, {{desc}}, {{transcript}}.

• YOUTUBE SUMMARIZER

  Improved YouTube video summaries command and format.