Real-time Compute Agent on Autopilot

Today’s top AI Highlights:

1. First real-time Computer Use agent that uses Your computer

2. AI pair programmer that deeply understands your codebase and learns as you work

3. Opensource multimodal embeddings model for visually-rich documents

4. Qwen 2.5-32B is the best open source OCR model

5. Connect any LLM to MCP tools with this MCP host desktop app

& so much more!

Read time: 3 mins

AI Tutorials

Voice is the most natural and accessible way for users to interact with any application and we see it being used the most for customer support usecases. But building a voice agent that can access your knowledge base can be complex and time-consuming.

In this tutorial, we'll build a Customer Support Voice Agent using OpenAI's SDK that combines GPT-4o with their latest TTS (text-to-speech) model. Our application will crawl documentation websites, process the content into a searchable knowledge base, and provide both text and voice responses to user queries through a clean Streamlit interface.

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.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to support us!

Latest Developments

Ace: The First Realtime Computer Autopilot 🤖🖥️🤲

Computer use AI agents that can autonomously complete multi-step tasks are the new paradigm of agentic automations, and there’s a new one here. Ace is a real-time computer use agent that performs tasks on your desktop using your mouse and keyboard, not on virtual machines or sandboxed environments. Check this thread out for some insane demos.

Ace can use all the tools on your computer, like right clicks and the search bar. And unlike other computer-use agents that are slow, would take minutes to complete a task, Ace does it like a superhuman. It’s incredibly fast! General Agents is currently offering a research preview, inviting people to test Ace.

Key Highlights:

1. Native System Execution - Ace operates directly on your computer, avoiding the performance overhead and resource limitations associated with virtualized environments. This native approach allows for faster task execution and more efficient use of system resources.

2. Behavioral Training - Unlike many VLMs, Ace employs a novel behavioral training approach. It learns from screen recordings, mouse movements, and keyboard inputs. This allows it to generalize better across diverse tasks.

3. Speed & Accuracy - Compared to models like OpenAI Operator, Anthropic's computer use agent, and other VLMs, Ace is fast and accurate. Benchmarks show that it is 20x faster in action prediction than OpenAI Operator and other agents. It also outperforms its competitors in left-click predictions with a stunning 95% accuracy.

4. Availability - General Computers is making ace-control models that power Ace available to selected partners through our developer platform. You can apply to get access to Ace’s research preview.

You’ve heard the hype. It’s time for results.

After two years of siloed experiments, proofs of concept that fail to scale, and disappointing ROI, most enterprises are stuck. AI isn't transforming their organizations — it’s adding complexity, friction, and frustration.

But Writer customers are seeing positive impact across their companies. Our end-to-end approach is delivering adoption and ROI at scale. Now, we’re applying that same platform and technology to build agentic AI that actually works for every enterprise.

This isn’t just another hype train that overpromises and underdelivers.
It’s the AI you’ve been waiting for — and it’s going to change the way enterprises operate. Be among the first to see end-to-end agentic AI in action. Join us for a live product release on April 10 at 2pm ET (11am PT).

Can't make it live? No worries — register anyway and we'll send you the recording!

Save your spot — register today

AI Coding Agent that Understands Large Codebases with 1000+ Files 🧠🗂️

Augment Code is the first code assistance platform built for professional software engineers working with large and complex codebases. It helps you understand code, debug issues, and ship faster with context-aware assistance.

The team has levelled up its platform with a new Augment Agent, your AI pair programmer that deeply understands your codebase, and learns as you work. This agent doesn't just suggest code - it writes it, runs it, logs every step, and can perform multi-file edits simultaneously. It cam understand 1000s of code files to deliver functional code that actually solves real engineering problems rather than creating new ones.

Key Highlights:

1. Large Context - With a 200K context window, Augment Agent can handle intricate codebases that cause other AI assistants to fail. It can edit multiple files across your codebase, create full PRs, and execute terminal commands while maintaining understanding of your project structure and conventions.

2. Persistent Memory - The agent learns your coding style, remembers previous refactors, and adapts to your infrastructure over time. This memory persists between sessions about your preferences or project specifics with each new task. It progressively becomes more aligned with how you work.

3. Integrated Development - It connects directly with your existing tools like GitHub for branch creation, commits, and PRs; integrates with project management systems like Linear, JIRA, and Notion; and works within VS Code and JetBrains IDEs.

4. Code Checkpoints and Safety - Every step the agent takes is logged and reversible with automatic code checkpoints. This gives you complete control over AI-generated changes without slowing down your workflow, essentially providing version control specifically for AI interactions.

5. Visual Debugging - Simply drag in a screenshot and the agent identifies UI issues (whether CSS, layout, or logic problems), suggests fixes, and runs only the relevant tests.

Augment Agent is available through a 14-day free trial with unlimited agent requests. Try it out before the offer goes away!

Quick Bites

Anthropic has launched Claude for Education, a specialized version of Claude designed specifically for higher education institutions. The new offering includes a unique “Learning mode” that guides students through reasoning processes rather than just providing answers. Anthropic is also releasing university-wide access agreements to make Claude available to all students.

OmniAI, a platform that turns unstructured documents into structured data, has released OCR benchmark results for the latest open source vision language models. In tests across 1,000 documents, Qwen 2.5 VL models (both 32B and 72B versions) achieved approximately 75% accuracy in JSON extraction tasks, matching GPT-4o's performance, even surpassing the purpose-built mistral-ocr model (72.2%). Surprisingly, Gemma-3 (27B) only reached 42.9% accuracy despite sharing architecture similarities with the top-performing Gemini 2.0 model.

Nomic has released Embed Multimodal, a new suite of embedding models that seamlessly process text, images, PDFs, and charts together. These models, available in 3B and 7B parameter sizes, outperform existing solutions when processing PDFs, charts, and research papers by simultaneously analyzing images and text without complex preprocessing pipelines. This can significantly improve RAG workflows by maintaining the visual context essential for technical documentation, financial reports, and product catalogs. Models are available on Hugging Face.

Tools of the Trade

1. MCP-Use: Open-source Python package for connecting any LLM to MCP tools in just 6 lines of code, without requiring desktop applications. It provides a straightforward client-agent structure for accessing MCP server capabilities from Python environments.

2. Dive: Open-source MCP Host desktop application that seamlessly integrates with any LLMs supporting function calling capabilities. It supports multiple AI models including OpenAI GPT, Claude, Gemini, and Ollama.

3. ToolJet: A low-code platform to build and deploy custom internal tools. It has a drag-and-drop app builder with 45 pre-built components to create complex applications in minutes. It connects to the most popular data sources and APIs out of the box.

4. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. It's honestly relieving to see that even the most powerful reasoning models get stuck during coding. I always assumed humans are just retards ~
   Tom Dörr
   
   

2. We know that OpenAI has built a GPT 4.5 version of the upgraded 4o model, which is already beating the "regular" 4.5 in many benchmarks. They are not releasing it publicly because it's too resource intensive.
   And we know that neither of those is a "reasoning" model, which would be even more resource intensive.
   But it's pretty clear that internally they have access to the "reasoning" version of those.
   Just think about it for a moment and draw the necessary conclusions. Take all the time you need. ~
   Bojan Tunguz

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!

Unwind AI - X | LinkedIn | Threads | Facebook

Awesome LLM Apps | Sponsor 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 😉