New on BuildTube: 2026-38

A skill you install into an AI coding agent so that "audit this codebase" runs a structured hunt for vulnerabilities rather than one opinionated read.

Install a ready-made routine so your AI coding agent plans before it builds, tests its changes and reviews its own work from a fresh context.

An inference engine written in plain C that tries to run very large models — hundreds of billions to trillions of parameters — on hardware that could not normally hold them.

Paste in text that sounds machine-written and get back a version that reads as if a person wrote it, with the same meaning.

A command-line tool for Git worktrees, aimed at people running several AI coding agents at once.

A free hands-on tutorial series on large language models, grown out of two courses at Shanghai Jiao Tong University on frontier language technology and AI security.

A local voice studio that covers both halves of speech: going out and coming in.

A desktop app that runs more than one AI coding agent at a time, each in its own separate copy of your project, so you can send the same request to all of them and keep the answer you like best.

A platform for building automations and AI agents by dragging steps onto a canvas and connecting them, with the option to drop into JavaScript or Python for the parts a visual editor handles badly.

A collection of ready-made AI personas: written instruction files (a persona is a character sheet that tells an AI how to behave) for specialists such as a front-end developer, a community writer, or someone who reality-checks your plan.

Set up Claude to help with one job, such as sales, legal, finance or customer support, using that role’s tools and shortcuts.

Give your AI coding agent the same engineering checks at every stage, from writing a spec to testing and shipping.

An operating layer installed on top of Hermes Agent, the open agent from Nous Research.

A development platform built around Postgres, the long-established open-source database.

A tool server that sits between a coding agent and the tools it calls, and keeps the bulky results out of the conversation.

Get AI review comments pinned to specific lines of your Git changes before a colleague looks at them.

An open-source coding agent available in five shapes from one engine: a VS Code extension, a JetBrains plugin, a terminal tool that also runs headless for automation, a desktop app, and a software development kit for building your own agents on top.

Anthropic’s coding agent for the terminal.

A Python engine that takes a starting piece of real information — a news item, a policy draft, a financial signal — and runs a simulated population of AI agents against it.

The Python library that defines how open models are written down, covering text, images, audio, video and combinations of them, for both running a model and training one.

Open a compiled program for Windows, macOS or Linux and read a reconstructed version of its code to work out what it does.

You load in documents and it builds a knowledge base you can ask questions of in plain language, plus an agent that works through multi-step questions and a wiki it keeps updated as the documents change.

A stock market tracking application you host yourself: live prices, personalised alerts and company detail pages, built as an open alternative to subscription market platforms.

A command-line tool and browser extension that connect an AI coding agent to your everyday browser, so it can work on sites you are already logged into instead of needing its own test accounts.

A self-hosted platform that gives each developer a workspace running on infrastructure you control — a cloud virtual machine, a Kubernetes pod, a Docker container — described in Terraform so the setup is written down rather than assembled by hand.

A distributed object storage system written in Rust, offering broad compatibility with Amazon S3’s interface so tools and applications that already speak S3 can point at it instead.

A complete book on building AI agents, published free and in the open by Bojie Li, organised around one formula: an agent is a language model plus context plus tools.

A private search engine for your own material.

A framework that puts a written specification between you and an AI coding assistant.

A keyboard launcher for macOS: one hotkey opens a search box that finds apps, files, clipboard history, snippets, emoji, calculator answers and window-arranging commands.

A shell tool that puts a single agent — the first mate — in front of a crew of autonomous coding agents.

Make your AI coding assistant reply with the next action first, numbered steps and no padding.

A local workspace that turns coding agents such as Claude Code, Codex, OpenCode or Cursor into research assistants: reviewing literature, proposing a hypothesis, changing code, launching an experiment and deciding what to try next.

An open business management platform that bundles the systems a small company normally buys separately: staff records and time tracking, a customer and sales pipeline, projects and tasks, invoicing and expenses, inventory, and an applicant tracking system for hiring.

Turn PDFs, Word files, slides, spreadsheets, images and audio into Markdown text that an AI model can read.

A command-line tool that boots a virtual iPhone on an Apple Silicon Mac using Apple’s own Virtualization framework together with the infrastructure Apple built for Private Cloud Compute research.

Run your draft through it and get back the same writing with more than 20 common AI-style patterns removed while your own voice stays.

Homebrew’s own macOS app for people who would rather click than type.

A self-hosted customer relationship system built around WhatsApp, with AI agents that answer, qualify and follow up on conversations inside it.

Use a spare iPhone, iPad or old Mac as a real second monitor for your Mac, dragging windows onto it like any other screen.

A free app for sharing files and messages with devices on the same local network — an alternative to AirDrop that works across Android, iOS, Windows, macOS and Linux instead of inside one company’s ecosystem.

A desktop workspace built specifically for AI coding agents, rather than an agent bolted into an editor or a terminal.

A modular operating system for logistics: dispatching jobs, managing a fleet, tracking vehicles live on a map, warehousing, order management and the finance around it, all in one platform you can host yourself.

A catalogue of skills — add-ons that teach an AI coding assistant a particular job — that the maintainers check before listing.

An open-source tool for open-source intelligence work: you build up an investigation as a visual graph, adding entities and the relationships between them, and extending it with your own collectors as you go.

An experimental preview of the architecture Alibaba intends to build its next model generation on, released with open weights.

Run Qwen3.8-27B, a 27-billion-parameter model that reads images and text, on your own computer at a file size you choose.

Run a language model that plans several steps, calls tools and works through long coding or research tasks, using a context of 256,000 tokens.

A cross-encoder — a model that reads two pieces of text together and judges their relationship — built by further training Qwen3.5.

Sort a document into fixed answers, such as a priority level or a yes-or-no flag, and get back valid JSON in one quick step on an Apple Silicon Mac.

Get answers from a 27-billion-parameter reasoning model that spends less time thinking to itself before replying.

Compressed builds of Qwen3.8-27B, Alibaba’s 27-billion-parameter model that reads images and video as well as text, packaged as GGUF files so they run through llama.cpp, Ollama or LM Studio on hardware that could not hold the original.

A small language model — 2.5 billion parameters in a standard Llama-style architecture — built specifically for running on a device rather than in a data centre, with a 131,072-token context that is unusual at this size.

A compressed, ready-to-run version of Swift-Qwen3.8-27B, which is Qwen3.8-27B adjusted to reach its answers with far less internal deliberation.

The Apple Silicon build of a heavily compressed 27-billion-parameter model, storing each weight as one of three values — minus one, zero or plus one — in an 8.6 GB download that includes a vision component.

Chat with a large language model on your own computer or phone, even when it has little memory to spare.

Run a small model on your own server that calls tools, writes code and follows instructions inside agent setups.