Short Stack #3Fri, Oct 16, 2026
01
Lovable apps can now call GPT-Live, Claude, Gemini and live-web tools
Lovable said on Oct 15 that apps you build in it can now use OpenAI's GPT-Live for real-time voice, the latest Claude models, Google's Gemini 4 Argon, Gemini Omni and Lyria, plus Perplexity, Firecrawl, Parallel and Jev by Typesafe AI. Describe the feature and Lovable picks a model or tool, or name one yourself in the chat.
⏱ 10-MIN TRYIn a Lovable project, prompt "Add a research panel that gives me a cited summary of my competitors. Use Perplexity." Then click through the citations to check they're real pages.
Source: lovable.dev
02
Perplexity open-sources SPACE, the sandbox behind Computer
On Oct 15 Perplexity released SPACE (Sandboxed Platform for Agentic Code Execution) under Apache 2.0. Each sandbox is a Firecracker microVM you drive over an HTTP API: create a Linux environment, run commands, stream output, pause and resume. Perplexity told Semafor nine frontier models failed to break out in 118 tries; that's the vendor's own test. Akamai, Crusoe and CoreWeave are lining up hosted options.
⏱ 10-MIN TRYOpen the repo, read the "Your first sandbox" walkthrough, and map its create/exec/delete calls against the sandbox your agent uses today.
Source: github.com
03
FLUX 3 Image Klein brings open-weights 4K editing to consumer GPUs
Black Forest Labs released FLUX 3 Image Klein on Oct 15: a 5.6B-parameter open-weights model for text-to-image, editing and multi-reference generation (up to 10 references), with native 4K and bounding boxes that pin where elements go. BFL says the INT8 build runs on 8GB RTX cards; its "ranks first" results are internal evals. Free commercial use is for companies under $5M revenue (FLUX Kommunity License).
⏱ 10-MIN TRYIn the BFL Playground or the Hugging Face Space, upload 3 brand references (logo, product shot, palette) and ask for one hero image that combines them.
Source: bfl.ai
04
Google moves inline links higher in AI Overviews and AI Mode
Google said on Oct 15 it is showing inline links higher in AI responses more often, adding thumbnails to end-of-response links, and testing a section that surfaces one or two official pages at the top of a response. It says Preferred Sources use is up more than 7x this year. Google shared no traffic numbers.
⏱ 10-MIN TRYSearch your top 3 buyer questions in AI Mode, note which sites get inline links, and make sure the page you'd want cited answers that question in its first paragraph.
Source: blog.google
05
ChatGPT can now read (and act on) your open Chrome tabs
With OpenAI's browser extension installed, you can tag Chrome in ChatGPT and it will use your open tabs as context, The Verge reported on Oct 15. In Work mode it can also fill forms, send messages, download files and group tabs. ChatGPT asks before using each new site by default, but the install prompt requests broad permissions, so read them first.
⏱ 10-MIN TRYOpen three tabs for tools you're comparing, tag @chrome in ChatGPT, and ask for a comparison table with pricing and source links. Stop it before it submits anything.
Source: theverge.com