- unwind ai
- Posts
- Open Agent Client Protocol
Open Agent Client Protocol
PLUS: xAI Grok Code Fast, New OpenAI GPT-Realtime
Shubham Saboo & Gargi Gupta
September 01, 2025
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
We have created a complete Google Agent Development Kit crash course with 9 comprehensive tutorials!
This tutorial series takes you from zero to hero in building AI agents with Google's Agent Development Kit.
What's covered:
Starter Agent - Your first ADK agent with basic workflow
Model Agnostic - OpenAI and Anthropic integration patterns
Structured Output - Type-safe responses with Pydantic schemas
Tool Integration - Built-in tools, custom functions, LangChain, CrewAI, MCP
Memory Systems - Session management with in-memory and SQLite storage
Callbacks & Monitoring - Agent lifecycle, LLM interactions, tool execution tracking
Plugins - Cross-cutting concerns and global callback management
Multi-Agent Patterns - Sequential, loop, and parallel agent orchestration
Each tutorial includes explanations, working code examples, and step-by-step instructions.
Everything is 100% Open-source.
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
CLI coding agents like Claude Code, OpenAI Codex, and Gemini CLI work brilliantly in the terminal, but a lot of developers still prefer using these tools directly within their IDE where they can see changes happen in real-time.
The problem is that each editor must build custom integrations for every agent they want to support, while agents must implement editor-specific APIs to reach users.
Zed just solved this with Agent Client Protocol (ACP) that standardizes communication between code editors (IDEs, text-editors, etc.) and coding agents. With ACP, any agent that speaks ACP can plug into a powerful UI, where the user can follow the agent around the codebase as it works, control its access to tools and MCP servers, and review all of its changes in a multi-buffer with full syntax highlighting and language server support.
The protocol is open-source under the Apache license. Gemini CLI in Zed is the first implementation of this protocol - editor maintainers can adopt the protocol to support the growing ecosystem of AI coding tools.
Key Highlights:
Structured Agent Communication - ACP uses JSON-RPC messaging through standard input/output streams, letting editors and agents exchange structured data instead of parsing messy terminal output or building one-off integrations for each tool.
Editor-Agnostic Design - The protocol separates agent logic from UI implementation, allowing the same agent to work across different editors while leveraging each editor's native capabilities like syntax highlighting and language server support.
MCP Server Integration - Agents can access MCP servers and tools through the editor's mediated permissions system, enabling controlled access to external services and data sources.
Extensible UI Primitives - ACP defines standardized message types for coding-specific interactions like patches, multi-file changes, and progress streaming, ensuring consistent user experiences across different agent implementations.
Speed-obsessed engineers at xAI decided that waiting 30 seconds for Claude or GPT to finish a coding task wasn't fast enough, so they built something that generates code at 160 tokens per second - fast enough to make you change how you work entirely.
Meet grok-code-fast-1, xAI's first dedicated coding model that's designed from scratch for developers who live in iterative loops of "code, test, fix, repeat."
Built on a completely new architecture and trained on a programming-heavy corpus, this model prioritizes blazing speed and low cost over perfect accuracy. The trade-off is intentional: xAI wants you to give it smaller, focused tasks and iterate rapidly instead of crafting massive prompts.
It's currently available free through launch partners like GitHub Copilot, Cursor, and Cline. The model achieved 70.8% on SWE-Bench-Verified and supports major languages, including TypeScript, Python, Java, Rust, C++, and Go.
Key Highlights:
Built for speed - The model generates code at up to 160 tokens per second with 90%+ cache hit rates, ideal for smaller, focused prompts instead of long complex ones.
Purpose-built architecture - Built from scratch with a new model architecture and trained on programming-heavy datasets including real-world pull requests and coding tasks, rather than being a general model adapted for code.
Agentic coding focus - Designed specifically for autonomous coding workflows where AI handles multi-step programming tasks with minimal supervision, including building projects from scratch and detailed bug fixes.
Availability - The model is available for free on GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf for a limited time. General API availability at $0.20/$1.50 per million input/output tokens plus $0.02 for cached tokens.
It’s go-time for holiday campaigns
Roku Ads Manager makes it easy to extend your Q4 campaign to performance CTV.
You can:
Easily launch self-serve CTV ads
Repurpose your social content for TV
Drive purchases directly on-screen with shoppable ads
A/B test to discover your most effective offers
The holidays only come once a year. Get started now with a $500 ad credit when you spend your first $500 today with code: ROKUADS500. Terms apply.
Quick Bites
Stop “Vibe Testing” Your AI applications
Google Labs just launched Stax, a new tool to rigorously test your AI stack, rather than getting into the endless loops of prompt tweaking to make the output feel better. You upload your prompts and expected answers, then Stax runs them through different models and grades the results using either built-in criteria (like is this factually correct?) or custom rules you define. Instead of manually checking if your chatbot responses sound right, Stax can test 100s of examples and give you scores to compare different approaches.
Claude Sonnet 4 and OpenAI GPT-5 in Apple Xcode
The new Xcode 26 beta brings Claude Sonnet 4 directly into Apple's development environment, joining the existing ChatGPT integration. You just need to connect your existing Claude account. OpenAI's GPT-5 is now the default ChatGPT model, accompanied by a "Reasoning" variant that deliberates longer on complex problems.
Save 20% on API cost with the new OpenAI GPT-Realtime
OpenAI has moved its Realtime API out of beta and launched GPT-Realtime, a more sophisticated speech-to-speech model than GPT-4o, that handles complex instructions with better precision. The new API now includes MCP server support, image inputs, and SIP phone calling capabilities for enterprise voice agents. GPT-Realtime can capture non-verbal cues like laughter, switch languages mid-sentence, and adapt to tone instructions like "speak quickly and professionally" or "speak empathetically in a French accent."
Two new exclusive voices (Cedar and Marin) with enhanced natural-sounding speech
Image processing capabilities enable voice agents to describe visual content during conversations
The API price for the new GPT-Realtime has been reduced by 20% - $32 / 1M audio input tokens and $64 / 1M audio output tokens.
Tools of the Trade
Vibecode Terminal - Use Claude Code, OpenAI Codex with GPT-5, Gemini CLI, and Cursor CLI in one platform. It runs these CLI agents in a web-based sandbox, so your personal computer and files are safe, and you can experiment freely.
web2mcp - Auto-generates MCP implementation for any web app. Stagehand crawls the app to map interactive elements, forms, and data collections, then GPT-5 converts this mapping into MCP configuration files. The resulting dynamic MCP server translates MCP calls into web application actions.
Claude Checkpoints - A version control tool that automatically monitors project files and creates snapshots when working with Claude Code for easy rollback. It has a built-in diff viewer to see exactly what changed between checkpoints
MCPcat - An open-source monitoring library for MCP server maintainers that gives logging and observability with a single line of code. It tracks user sessions, tool calls, and agent behavior with out-of-the-box integration with existing platforms like Datadog and Sentry.
Awesome LLM Apps: A curated collection of LLM apps with RAG, AI Agents, multi-agent teams, MCP, voice agents, and more. The apps use models from OpenAI, Anthropic, Google, and open-source models like DeepSeek, Qwen, and Llama that you can run locally on your computer.
(Now accepting GitHub sponsorships)
Hot Takes
posted about this a while ago but even more convinced of it now
i think LLMs are going to plateau right at the point where they make programmers lives even better than they already were but fail to eliminate them ~
daxif i worked on nano banana and found out marketing renamed it to Gemini-2.5-Flash-Image-Preview i would quit ~
vik
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 😉
1
Reply