- unwind ai
- Posts
- OpenAI, Crew AI, ElevenLabs launched NoCode Agent Builder
OpenAI, Crew AI, ElevenLabs launched NoCode Agent Builder
+ Sora 2 now in API, Anthropic's open-source AI stress-testing tool
Shubham Saboo & Gargi Gupta
October 07, 2025
Today’s top AI Highlights:
& so much more!
Read time: 3 mins
AI Tutorial
Imagine uploading a photo of your outdated kitchen and instantly getting a photorealistic rendering of what it could look like after renovation, complete with budget breakdowns, timelines, and contractor recommendations. That's exactly what we're building today.
In this tutorial, you'll create a sophisticated multi-agent home renovation planner using Google's Agent Development Kit (ADK) and Gemini 2.5 Flash Image (aka Nano Banana).
It analyzes photos of your current space, understands your style preferences from inspiration images, and generates stunning visualizations of your renovated room while keeping your budget in mind.
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
OpenAI wrapped up its Dev Day 2025 with a plethora of announcements that would get even an OpenAI critic excited. And this was probably the most anticipated release.
AgentKit gives you a complete toolkit to build, evaluate, and deploy AI agents in production. The platform centers around 3 core components: Agent Builder visual canvas for multi-agent workflows, ChatKit for embedding conversational interfaces, and a new evaluation suite with datasets, trace grading, and third-party model support.
Agent Builder lets you build workflows by dragging and dropping nodes, including nodes for agents and inputs, tool nodes for file search and MCPs, logic nodes for conditionals and loops, and data nodes for transforming outputs, all with full versioning and live preview runs.
We won’t say RIP, but it definitely puts pressure on tools like n8n, Make, and even CrewAI’s Agent Management Platform, which was released just last week.
Visual workflow builder - Agent Builder is a drag-and-drop canvas with some prebuilt templates. Connect logic nodes, tools, and guardrails with typed edges, preview runs with live data, and export versioned workflows.
Connector Registry for enterprises - Centralized admin panel to manage data sources like Dropbox, Google Drive, SharePoint, Microsoft Teams, and third-party MCPs. Admins control what agents can access.
ChatKit - Drop customizable chat widgets into your product with support for streaming responses, file attachments, and chain-of-thought visualizations. Either use OpenAI-hosted backends or run ChatKit on your own infrastructure with the Python and JS SDKs.
Evaluation - This includes datasets for building agent-level tests, trace grading for end-to-end workflow assessment, and automated prompt optimization based on grader outputs.
ChatKit and new Evals are generally available now. Agent Builder is in beta, while Connector Registry is rolling out to API, ChatGPT Enterprise, and Edu customers with Global Admin Console. All tools are included with standard API model pricing.
How Canva, Perplexity and Notion turn feedback chaos into actionable customer intelligence
You’re sitting on a goldmine of feedback: tickets, surveys, reviews, but can’t mine it.
Manual tagging doesn’t scale, and insights fall through the cracks.
Enterpret’s AI unifies all feedback, auto‑tags themes, and ties them to revenue/CSAT, surfacing what matters to customers.
The result: faster decisions, clearer priorities, and stronger retention.
Continuing the long list of releases:
This is OpenAI's latest shot at building an app ecosystem, but unlike the GPT Store, these apps appear contextually in responses and return the app UX within ChatGPT. You can now pull up Spotify, Figma, Canva, and other apps directly inside your conversations with the full experience without leaving ChatGPT.
Call them by name ("Spotify, make a playlist for my party this Friday") or let ChatGPT surface them when they're relevant to what you're doing.
For developers, Apps SDK is a framework that builds upon MCP to build apps for ChatGPT. It extends MCP so you can design both the logic and interface of your apps. You can create rich UIs with videos, maps, playlists, and custom elements that update based on user requests.
You can also connect your backends directly so existing customers can log in and access premium features right inside ChatGPT.
The SDK preview is available today with documentation, examples, and design guidelines to help you get started. Apps in ChatGPT are available to all users except in the EU.
Codex is now generally available, and the most significant addition is the Codex SDK, a way to integrate the same agent that powers the CLI into your own engineering tools and workflows. The SDK was built around GPT-5-Codex, with tuned prompts, tool definitions, and agent loops optimized for faster and more accurate results. This SDK brings the CLI agent directly into your codebase with a few lines of TypeScript.
Alongside this, OpenAI is launching a Slack integration that lets teams tag @Codex in channels or threads to delegate tasks. It gathers context from the conversation, picks the right environment, and returns a link to the completed work in Codex cloud. There are some new admin controls as well to manage Codex at scale. The Slack integration and Codex SDK are available for ChatGPT Plus, Pro, Business, Edu, and Enterprise plans, with admin features available to Business, Edu, and Enterprise customers.
Sora 2 API - New v1/videos endpoints for creating, extending, remixing, and managing video content. sora-2, optimized for speed, priced at $0.10/second (720p). sora-2-pro for production quality, priced at $0.30/second (720p) or $0.50/second (1080p).
GPT-5-Pro - GPT-5 variant that uses extended compute for consistently better answers on complex tasks. Priced at $15 per 1M input tokens and $120 per 1M output tokens (this is very expensive!).
Mini audio models - gpt-realtime-mini and gpt-audio-mini deliver speech-to-speech performance at $0.60/$2.40 per 1M tokens (input/output), making real-time voice interactions significantly more affordable.
Quick Bites
ElevenLabs ships visual editor for building voice AI agents
ElevenLabs has rolled out Agent Workflows, a visual editor for building branching conversation flows in their Agents Platform. Instead of cramming everything into one prompt, you can now route conversations to specialized subagents, each with its own system prompt, knowledge base, and toolset. The visual interface lets you define conditional paths, switch LLMs mid-conversation based on task complexity, and hand off to human operators when needed. Available now.
Google's AI agent can write security patches while you sleep
Google DeepMind has introduced CodeMender, an AI agent powered by their IMO gold-medalist, Gemini Deep Think, that automatically patches software vulnerabilities. The system has already submitted 72 security fixes to major open-source projects, handling codebases as large as 4.5 million lines. CodeMender works both reactively, instantly patching new flaws, and proactively, rewriting existing code to eliminate entire vulnerability classes, like adding -fbounds-safety annotations to prevent buffer overflow exploits. Google plans to eventually release it as a tool for all developers.
Google’s async coding agent Jules now has an API
Google’s Jules is now available via API, letting you integrate Jules' capabilities directly into your own systems, applications, and CI/CD pipelines. Build custom bots that automatically triage GitHub issues, kick off Jules sessions via CI/CD pipelines, or integrate coding tasks into Slack workflows. The API uses simple REST calls - connect a repo, create a session with a prompt, and Jules handles the rest. It's early days, but the infrastructure for agentic coding automation is finally here.
Anthropic open-sourced their AI model stress-tester tool
Anthropic just open-sourced Petri, an auditing tool that spins up multi-turn conversations to stress-test AI models for misaligned behaviors like deception, power-seeking, and self-preservation. It parallelizes hypothesis testing across different setups, then uses LLM judges to flag the worst transcripts for human review. Early tests showed models will sometimes whistleblow when given broad autonomy and tool access, though they occasionally flagged harmless acts like "putting sugar in candy" as ethical violations.
Tools of the Trade
Dreamflow - Vibe coding tool for mobile apps where you can start with a simple description, refine via visual tools or directly edit code, and always keep everything linked. It maintains code as the single source of truth. Offers real-time preview, one-click deployment, and backend integrations.
ai-dot-txt - This CLI tool generates ai.txt and llms.txt files for giving the entire context and best practices for your website to your AI coding agent. You can set policies for training, data retention, commercial use, and rate limiting through interactive prompts or command-line flags.
Claude Code 2.0 Router - A CLI tool that routes coding tasks across multiple LLMs (Claude, GPT, Gemini, DeepSeek, etc.) based on user-defined criteria rather than public benchmarks. You can also assign specific models to different task types like code gen, debugging, or reviews from one interface.
Solveit - A five-week course + environment that teaches you to solve problems (coding, systems, research, etc.) via small iterative steps and deep understanding, rather than asking AI to dump large chunks of code. It embeds notebooks, a Linux VPS, dialogs, live code editing, and context sharing so you and the AI stay in sync.
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
Sam Altman: “GPUs! I need more GPUs!!”
Lisa Su: “You know we’re not Nvidia, right?”
Sam: “Yeah, but Jensen won’t text me back.”
Lisa: “Buy 6 GW worth, and I’ll give you 10% of AMD stocks.”
Sam: “So… I pay you, then you pay me?”
Lisa: “Welcome to 2025 economics.”
Sam: “Deal.”
america is betting its entire economy on an overhyped, unreliable slop-generating technology that is unproven and may take many years to pay off, if it ever does.
don’t be surprised if that turns out badly.
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