- unwind ai
- Posts
- Agentic Browser that Sees All Your Tabs
Agentic Browser that Sees All Your Tabs
PLUS: Deep Research in ChatGPT Projects, Open standard for stateful AI agents
Today’s top AI Highlights:
This is the Browser War and AI agents are the weapon
Voice-activated AI teammate that listens to your entire meeting
Pack your entire AI agent into one portable file
Run Deep Research on your own files in ChatGPT Projects
AI Sheets turns any spreadsheet into an AI-powered research machine
& so much more!
Read time: 3 mins
AI Tutorial
Building good research tools is hard. When you're trying to create something that can actually find useful information and deliver it in a meaningful way, you're usually stuck cobbling together different search APIs, prompt engineering for hours, and then figuring out how to get the results into a shareable format. It's a headache, and the results are often inconsistent.
In this tutorial, we'll build an AI Domain Deep Research Agent that does all the heavy lifting for you. This app uses three specialized agents that are built using the Agno, Qwen 3 235B model via Together AI, and use tools via Composio to generate targeted questions, search across multiple platforms, and compile professional reports.
What makes this deep research app different from other tools out there is its unique approach: it automatically breaks down topics into specific yes/no research questions, combines results from both Tavily and Perplexity AI for better coverage, and formats everything into a McKinsey-style report that's automatically saved to Google Docs.
We share hands-on tutorials like this every week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.
Latest Developments
Your meetings just got their own always-on (and much smarter) Siri. Fireflies has launched Talk to Fireflies, a voice-activated meeting assistant that listens throughout your calls and responds instantly when you say "Hey Fireflies."
Whether you need to recall what someone said 20 minutes ago or pull the latest market data from the web, this AI teammate delivers answers without making you pause the conversation or juggle multiple apps.
Just say "Hey Fireflies" and ask anything from "who mentioned the budget concerns?" to "what are the latest trends in our industry?" It seamlessly blends your live conversation data with Perplexity's web search, giving you the context you need to make faster decisions right inside your calls.
Key Highlights:
Always-On Intelligence - The assistant continuously processes your meeting transcript, ready to answer questions about speakers, decisions, action items, or timestamps whenever you need them.
Web-Enhanced Responses - Through Perplexity integration, get real-time market data, news, and research findings directly in your meeting without switching tabs or apps.
Smart Task Management - Ask Fireflies to assign action items, tag owners, and bookmark important moments during the call for seamless follow-up.
Cross-Platform Access - Available on Google Meet, Teams, Zoom, and other major platforms, with free access for all users and premium partnership benefits for paid subscribers.
Every AI company is scrambling to build browsers faster than Chrome can integrate Gemini, and The Browser Company just launched theirs first. Dia is an AI browser that makes AI conversation the default way to interact with the web.
The team built Dia around a simple premise: what if your browser had an AI assistant that could see and remember everything you do online? Dia combines Chrome's familiar interface with a ChatGPT-like sidebar that accesses all your tabs, browsing history, and logged-in accounts to provide contextual help without the usual copy-paste friction.
Key Highlights:
Universal browser access - The AI can read content from all your logged-in accounts and services, eliminating the need to manually share information between different tools and websites.
Context-aware assistance - Rather than starting fresh each time, Dia's AI builds on your browsing patterns and history to provide increasingly personalized help with tasks.
Specialized AI skills - The browser automatically routes different requests to purpose-built AI capabilities, so shopping queries get different treatment than coding questions.
Beta access - All existing Arc members get immediate access to Dia, with current beta users able to send invites to others as the company tests its AI-first approach.
Building AI agents usually means starting from scratch every time, but what if you could save, share, and clone your smartest agents instantly?
This is a big pain point in AI agent development: each framework stores agent data differently, making it impossible to move agents between systems or collaborate. Letta released Agent File (.af), the first open standard file format for serializing stateful AI agents with persistent memory and behavior.
.af file packages system prompts, editable memory blocks, tool configurations, and LLM settings into a single standardized format that works across multiple frameworks.
Think of it as a save file for your AI agents that captures everything - their personality, memories, tools, and learned behaviors - in one package you can share across frameworks.
Key Highlights:
Complete Agent State Capture - Agent Files contain everything needed to recreate an identical agent: message history, memory blocks, tool definitions with source code, environment variables, and model configurations. You get the exact same agent behavior when importing the .af file anywhere.
Framework Portability - Originally designed for Letta, the .af format is an open standard that other frameworks can adopt by mapping components to their equivalent features. This enables true agent portability across different environments and systems.
Ready-to-Use Agent Library - Letta has given downloadable example agents including MemGPT researchers, deep research agent, and workflow automation agents to jump-start your projects.
Version Control - The standardized format enables proper versioning of agent states, making it easy to track changes, roll back to previous versions, and collaborate on agent improvements.
Quick Bites
Hugging Face just launched ScreenSuite, a new benchmark suite focused entirely on Computer-use AI agents. It’s designed to test how well these agents perform across a full range of tasks like reading screen info, clicking accurately, and solving multi-step workflows on systems like Windows, Android, and the web. What makes ScreenSuite stand out is that it sticks to vision-only inputs - no behind-the-scenes help from DOM or accessibility trees, making it a tough and realistic testbed.
You can try it locally in under a minute, and it’s already being used to benchmark top VLMs like Qwen-2.5-VL, GPT-4o, Holo1, and UI-Tars.
OpenAI just gave a solid upgrade to Projects in ChatGPT, making them more useful for complex, focused work. You can now run deep research with in-depth, multi-step research on your chats, files, and notes. Voice mode is also supported inside projects, so you can talk through ideas hands-free. Plus, mobile users get new capabilities: upload files, switch models, and enjoy better memory that connects past chats within a project.
Tools of the Trade
Code Graph MCP Server: Create knowledge graphs for navigating and understanding massive, legacy, or poorly documented repositories, and use these knowledge graphs across MCP clients like Cursor and Claude.
AI Sheets: A surprisingly powerful and free AI-enhanced spreadsheet tool. Create, analyze, and automate spreadsheet enrichment and dataset creation using open LLMs. You can bring in an existing spreadsheet, and it can identify existing data and build on it. It can also search the web to fill with up-to-date data.
Markdown Rules MCP Server: Opensource tool that converts standard Markdown docs into intelligent context for AI coding agents, allowing you to define rules and context that work across multiple AI tools instead of being locked into Cursor's format.
Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, MCP, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.
Hot Takes
DeepSeek-V4/R2 will crush american labs if they don't get crazy and go for 90% margins like Anthropic ~
Lisan al GaibI think the most dangerous skillset combo right now is builder (designer or engineer) AND filmmaker. You're a 1 person factory from creation to distribution. ~
GREG ISENBERG
That’s all for today! See you tomorrow with more such AI-filled content.
Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!
PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉
Reply