Gemini Omni 3.5 demos, Vercel Docker sandboxes, and Product Hunt agent tools
TL;DR
Google showed Gemini Omni and 3.5 in action while Vercel expanded sandbox tooling and new agent products appeared on Product Hunt and Hugging Face.
What shipped
On 29 May Google released demo videos for Gemini Omni and Gemini 3.5 while Vercel added Docker and port controls to its sandboxes. Product Hunt featured eight new AI tools aimed at agents and workflows. Two image-text models also trended on Hugging Face.
Vendor launches
Google focused on Gemini Omni and Gemini 3.5 with eleven videos and an AI-built quiz. Vercel released three sandbox updates that let agents run Docker containers and open port 8,080 while also publishing guidance on stopping inference theft.
- •Inference theft protection Vercel described how exposed AI endpoints can face attacks that generate bills of thousands of dollars from single expensive prompts and shared steps to lock them down.
- •I/O 2026 quiz Google used its AI Studio to build a quiz covering the main I/O 2026 announcements in a few minutes of prototyping.
- •Gemini Omni and 3.5 demos Google published eleven videos that walk through real tasks performed by Gemini Omni and Gemini 3.5 after the I/O event.
- •Docker in Vercel Sandbox Vercel now lets agents install Docker, run containers, and test services such as Redis or Postgres inside isolated sandboxes.
- •Waterloo Futures Lab University of Waterloo students built working AI prototypes including a sign language tutor that targets education and workplace use cases.
- •Port 8,080 in Sandboxes Vercel opened port 8,080 for ingress in sandboxes so developers can test web services without changing the controller port.
Product Hunt picks
Eight new tools launched on Product Hunt that target design, tracing, frontend bridging, marketing, video editing, usage tracking, web monitoring, and repository memory for agents.
- •Drafted Drafted generates home designs in seconds so users can preview layouts without hiring a designer.
- •PromptLayer PromptLayer records every AI request and workflow in one timeline so teams can review costs and debug runs.
- •Vibeocus Lens Vibeocus Lens links a live frontend page straight to an AI agent for instant testing and control.
- •Agent A by Ahrefs Agent A pulls Ahrefs SEO data to run marketing tasks such as keyword research and content planning.
- •Clipline Clipline cuts long videos into Shorts, Reels, or TikToks inside Telegram for fast social posting.
- •TrackNotch TrackNotch shows live LLM token counts and costs inside the Mac notch for constant visibility.
- •/monitor by Firecrawl Firecrawl monitor sends alerts to agents when chosen web pages change so workflows stay current.
- •GPS GPS stores repository context so coding agents recall past files and decisions across sessions.
Hugging Face trending
Two new image-text-to-text models appeared in the Hugging Face trending list and target multimodal tasks such as OCR and visual question answering.
- •Step-3.7-Flash stepfun-ai released Step-3.7-Flash, an image-text-to-text model that supports multimodal inference and fine-tuning on the Hub.
- •PaddleOCR-VL-1.6 PaddlePaddle released PaddleOCR-VL-1.6, an image-text-to-text model focused on OCR and document understanding tasks.
What this means for you
For Vibe Builders: You can now test Gemini 3.5 capabilities directly from the new demo videos and spin up Docker containers inside Vercel sandboxes to run test services without local setup. Product Hunt tools such as Vibeocus Lens and GPS let you connect live frontends to agents and give them repo memory in minutes. These releases cut the time between idea and working prototype.
For Non-techies: Google's Gemini demos show what the newest models can do for everyday tasks like quizzes and sign-language practice. Tools on Product Hunt such as Drafted for home design and Clipline for quick video clips let small teams ship simple AI projects today without writing code. Watch for usage trackers like TrackNotch if you start running many prompts.
For Developers: Vercel added Docker and port 8,080 support so agents can run containerized dependencies like Postgres during tests. PromptLayer and Firecrawl monitor give production teams timelines and change alerts for agent workflows. The two trending Hugging Face models offer lightweight multimodal options you can benchmark before adding them to pipelines.
What to watch next
Track follow-up posts from Google on Gemini Omni fine-tuning access. Watch for any new Vercel sandbox limits after the Docker update. Check whether the Product Hunt agent tools release public benchmarks this week.
Harsh’s take
Most of the day focused on sandbox convenience and agent visibility rather than raw model gains. The inference-theft post from Vercel stands out because it shows real cost risk once agents run in public. Builders should test one sandbox feature and one Product Hunt tool side by side this week to measure actual time saved versus added monitoring overhead.
by Harsh Desai
Sources
Vendor launches
- •Protecting against inference theft
- •Take our I/O 2026 quiz, vibe coded in Google AI Studio.
- •11 demos of Gemini Omni and Gemini 3.5 in action
- •Run Docker containers inside Vercel Sandbox
- •Check out real-life AI prototypes from the Futures Lab.
- •Port 8,080 is now available in Vercel Sandboxes
Product Hunt picks
Hugging Face trending
More AI news
- Weekly DigestClaude Code adds Opus 4.8 and background agents, Codex refines TUI, Cursor holds steady
Claude Code rolled out Opus 4.8 support, background shell sessions, and skill reload tools while Codex focused on TUI fixes and permission profiles with no movement from Cursor.
- Daily RoundupOpus 4.8's coding jump, Gemini 3.5 Flash, and the agent tools you can run today
Fresh models on Hugging Face and Replicate plus Vercel infrastructure updates give builders quicker paths to ship agents and image tools while Google I/O and Product Hunt releases add new options for daily workflows and production stacks.
- Daily RoundupNVIDIA Pushes AI Factories While Agents Gain Calling and Local Stacks
NVIDIA framed AI factories around token economics, Google and Vercel refined search and deployment views, and eight Product Hunt tools added PDF parsing, phone calls, local LLM stacks, and code validation for agents.