Meet HoloTab by HCompany. Your AI browser companion.

Source: Hugging Face - Blog | Tags: ai, automation, chrome-extension | Published: 2026-04-15 | Novelty: 51%

HCompany has released HoloTab, a Chrome extension that uses the advanced Holo3 AI model to automate web-based tasks with zero setup required. Users can record their routines and then run them anytime without technical skills, making complex web tasks like cross-referencing data or filtering job listings more efficient. The tool is free and accessible at https://chromewebstore.google.com/detail/holotab/hlaoiikljjgcjdhkakedfngifaopbcop.

Gemini 3.1 Flash TTS

Source: Simon Willison's Weblog | Tags: accent, gemini, prompting, text-to-speech, tts | Published: 2026-04-15 | Novelty: 43%

The article introduces Gemini 3.1 Flash TTS, a new text-to-speech model that supports prompting for specific audio outputs. Examples provided include prompts for different accents and styles, such as 'Jaz R.' from Brixton with an energetic Top 40 radio style, and variations for Newcastle and Exeter accented voices. The examples demonstrate the model's ability to produce varied speech based on detailed directorial notes.

datasette PR #2689: Replace token-based CSRF with Sec-Fetch-Site header protection

Source: Simon Willison's Weblog | Published: 2026-04-14 | Novelty: 39%

{ "summary": "datasette PR #2689 implements a new CSRF protection mechanism inspired by Go 1.25 and Filippo Valsorda's research, replacing the previous token-based approach with Sec-Fetch-Site headers, thus eliminating the need for <input type="

Zig 0.16.0 release notes: "Juicy Main"

Source: Simon Willison's Weblog | Tags: dependency-injection, main-function, zig | Published: 2026-04-15 | Novelty: 38%

Zig 0.16.0 introduces 'Juicy Main', a dependency injection feature for the main() function, enhancing accessibility to various utilities such as allocators and environment variables through the std.process.Init parameter. This update significantly streamlines program initialization by directly providing necessary resources within the main function scope.

Inside VAKRA: Reasoning, Tool Use, and Failure Modes of Agents

Source: Hugging Face - Blog | Tags: benchmark, multi-hop-reasoning, policies, tool-selection | Published: 2026-04-15 | Novelty: 38%

VAKRA introduces a benchmark for evaluating multi-hop reasoning and tool selection under policy constraints, where GPT-OSS-120B outperformed other models due to better understanding of complex schemas. The dataset comprises four capabilities: Business Intelligence APIs (SLOT-BIRD and SEL-BIRD), Multi-Hop Reasoning, Multi-Hop Multi-Source Reasoning, and Tool-Use Policies. Notably, Gemini-3-flash-preview excelled in the Dashboard API capability but struggled with synthesizing correct answers from tool outputs.

Quoting Kyle Kingsbury

Source: Simon Willison's Weblog | Tags: accountability, jobs, ml | Published: 2026-04-15 | Novelty: 36%

Kyle Kingsbury suggests that accountability for machine learning systems will be a new job, with humans acting as 'meat shields' to absorb the blame when these systems misbehave. He envisions various forms of responsibility, from internal reviews by Meta to external legal penalties and formalized roles like Data Protection Officers. The metaphor of 'throwing under the bus' emphasizes the potential for outsourcing accountability.

Quoting John Gruber

Source: Simon Willison's Weblog | Tags: app-store, apple, developers, software-quality | Published: 2026-04-15 | Novelty: 35%

According to John Gruber, the primary advantage of Apple's App Store is the high quality of apps available on its platforms, which drives user adoption across iPhone, Mac, and iPad. However, this edge is diminishing as fewer developers are motivated to create well-crafted native applications for these devices, leading to a regression in the quality of software.

Trusted access for the next era of cyber defense

Source: Simon Willison's Weblog | Tags: cybersecurity, gpt, openai | Published: 2026-04-14 | Novelty: 25%

OpenAI has introduced GPT-5.4-Cyber, a variant of their model fine-tuned for cybersecurity use cases. They have also launched the Trusted Access for Cyber program, which allows users to verify their identity via Persona and gain reduced friction access to OpenAI's security tools. This move aims to democratize access but still requires an extra application process for some of the best security features.

Gemini 3.1 Flash TTS

Source: Simon Willison's Weblog | Tags: gemini, text-to-speech, tts | Published: 2026-04-15 | Novelty: 18%

This article by Simon Willison, posted on April 15, 2026, discusses Gemini 3.1 Flash TTS, focusing on a specific beat or development in text-to-speech technology within the Gemini project. While no new ideas or unique points of view are explicitly detailed, the post highlights the growing importance and advancements in TTS technologies under the Gemini umbrella.

datasette-ports 0.3

Source: Simon Willison's Weblog | Tags: database, datasette, path, ports | Published: 2026-04-15 | Novelty: 15%

The update to datasette-ports 0.3 now includes the full path to each database file and shows working directories derived from each PID, providing a detailed view of running Datasette instances on the user's laptop.