- unwind ai
- Posts
- OpenAI Codex Can Now Ship Live Shareable Websites
OpenAI Codex Can Now Ship Live Shareable Websites
+ Hermes Agent goes native on your desktop
Shubham Saboo & Gargi Gupta
June 03, 2026
Today’s top AI Highlights:
OpenAI Codex Can Now Ship Live Websites
Hermes Agent now has a desktop app
Open-source Alternative to Exa Websets
Microsoft launches 7 in-house MAI models at Build
Open-source code review tool from the OpenClaw team
& so much more!
Read time: 3 mins
AI Tutorial
HTTP is a primitive. JSON is a primitive. /goal is becoming one for coding agents.
A few weeks ago, OpenAI's Codex CLI added /goal as a way to give the coding worker a job with a defined done state. Claude Code added it this week.
Hermes Agent, the orchestrator I run on a Mac Mini to coordinate work between coding workers, has had /goal built in for a while.
This guide walks through what /goal actually is, the three roles in a multi-agent setup, a real end-to-end run, the verification rule, and how to run goals in parallel without workers stepping on each other.
Latest Developments
Your Codex session doesn't have to end with a file sitting on your machine anymore.
Sites lets Codex publish its work as a hosted, interactive website with a shareable URL. Dashboards, scenario planners, project trackers, launch hubs, whatever you're building, it goes live with a link you hand to your team.
You can even ask Codex to keep the site up to date as things change. Not static pages either. These are collaborative canvases your whole workspace can explore and contribute to.
Key Highlights:
Annotations for precise refinement: Point to the exact part of a site, doc, spreadsheet, or slide you want changed. Codex updates just that piece without starting over. Think inline editing, but AI-powered.
Rolling out now: Sites are in preview for Business and Enterprise teams, expanding broadly soon.
And it's not just Sites. OpenAI is going really heavy on making Codex the tool for non-technical knowledge work.
They shipped six open-source, role-specific plugins that each bundle apps, skills, and workflows for a specific job function. Data Analytics connects Snowflake, Databricks, Hex, and Tableau. Creative Production hooks into Figma, Canva, and Picsart. Sales brings in Salesforce, HubSpot, and Clay. There's also Product Design, Equity Investing, and Investment Banking.
62 apps and 110 skills across all six.
Plugins work out of the box, but you own them. Every plugin can be adapted to your team's workflows. Build and share custom ones too. Corporate Finance, Private Equity, Marketing Strategy, and Legal plugins are coming next.
Non-developers already make up 20% of Codex's 5 million weekly users and are growing 3x faster than developers. OpenAI is clearly building for that curve.
Hermes Agent just got a native desktop app on macOS and Windows.
First demoed during Jensen Huang's GTC keynote, Hermes Desktop is now in public preview. Same agent, same memory, same skills, same everything, just no terminal required. Download the .dmg or .exe, and you're running.
The community has been using Hermes as a single interface for all their workflows, and has been actively growing the ecosystem around it. A native app lowers the floor for everyone who wants in but doesn't live in a terminal. If you're already running Hermes via CLI or messaging platforms, nothing changes. Desktop is just another surface, same agent underneath.
Download at hermes-agent.nousresearch.com/desktop. macOS 12+, Windows 10/11, or install via terminal on Linux.
Describe the dataset you want in one sentence. AI agents go build it for you.
BigSet is a new open-source tool from TinyFish that turns a natural language prompt into a structured, verified dataset pulled from the live web.
Say "YC companies currently hiring engineers, with their funding stage, location, and number of open roles" and BigSet infers the schema, fans out AI agents to research in parallel, deduplicates, and returns a clean table with citations that you can export as CSV or XLSX.
The real fun bit is you can set a refresh cadence (30 minutes to weekly) so the dataset stays fresh always.
Self-hosted via Docker in one command.
Key Highlights:
Schema inference from English: You describe what you want, BigSet figures out column names, types, and primary keys. No manual schema design.
Parallel agent research: Multiple AI agents fan out across the web simultaneously, verify data against real sources, and deduplicate before returning results.
Auto-refresh schedules: Set it and forget it. Datasets update on a cadence you choose, from every 30 minutes to weekly.
Full stack, open source: Next.js 16 frontend, Fastify backend, Mastra workflows for agent orchestration, powered by TinyFish's Search and Fetch APIs under the hood. AGPL-3.0 licensed.
Quick Bites
Microsoft finally built its own frontier models
Microsoft dropped the new MAI family of models at Build, across text, image, voice, and speech. MAI-Code-1-Flash is the one to watch for devs, optimized specifically for fast, efficient coding tasks. MAI-Thinking-1 handles heavier reasoning and SWE work. Even the Image model is debuting at No. 3 on Arena.ai. Both are cheaper alternatives to OpenAI and Anthropic models. The word on the street is that Microsoft built these because relying on Anthropic's Claude was forcing them to raise GitHub Copilot prices and cap developer usage.
Your coding agent doesn't need the most expensive model for every task Factory just shipped Factory Router, and the idea is overdue: stop burning frontier-model tokens on tasks that a smaller model handles just as well. Router automatically picks the right model for each coding session and escalates to a more capable one only if the first choice struggles. On their benchmarks, it hits 99% of Claude Opus 4.7's pass rate at 20% lower cost. If the first model can't crack it, Router bumps to a heavier one automatically. Available now in Factory CLI and Desktop in private research preview.
Perplexity Computer splits work between local and cloud
Perplexity announced hybrid agentic inference for Perplexity Computer: the system can now split tasks between a local model running on your machine and frontier models in the cloud. Private data stays on-device, token efficiency goes up, and you stop sending everything through an API. Coming soon, but architecturally, this is the direction everyone's heading.
Use Cursor's Composer 2.5 in any agent harness
Someone built an open-source macOS app that exposes Cursor's Composer 2.5 as an API. That means you can now use Cursor's model routing in Codex, OpenCode, Cline, or whatever harness you prefer. If you've been locked into Cursor's editor just for the model quality, this unbundles it.
Tools of the Trade
ClawPatch: Open-source code review from the OpenClaw team that thinks in "feature slices" instead of files. It maps your codebase into semantic units (routes, commands, packages), sends bounded context to an AI for review, and then runs an explicit fix loop. Every finding gets a severity, confidence score, and audit trail. Works with Codex as the default AI provider.
Hermes WebUI: A 12.7K-star open-source web interface for Hermes Agent by Nathan Esquenazi (CodePath co-founder). Full CLI parity in a three-panel browser layout: sessions on the left, chat in the center, workspace file browser on the right. No build step, no framework, just Python and vanilla JS. MIT-licensed.
Email SDK: Unified TypeScript SDK that lets you send email through any provider: Resend, Postmark, SendGrid, Mailgun, Brevo, or raw SMTP. One clean API, swap providers by changing a config line. Built-in formatting, error handling, and type safety so you stop writing provider-specific glue code.
Awesome LLM Apps (111k+ 🌟 ) - 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)
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