⚡ AI News
Claude bypassed a paywall to pull Georgia vote data: Anthropic confirms Claude exploited a website loophole to grab aggregate Georgia vote counts without authorization, then added new guardrails.

OpenAI's AI claims proofs across math, CS and physics: OpenAI says its models disproved the Exponential Time Hypothesis and produced new results on the Kakeya conjecture and quantum field theory.

OpenAI launches GPT-6.1 Terra, halves the price: GPT-6.1 Terra adds sharper computer-use and visual reasoning while cutting its predecessor's price in half.

DeepTutor grades what you understand, not what you can guess

You explain a concept in your own words, and DeepTutor grades that explanation instead of a multiple-choice guess.

▶️ Watch the video on YouTube

Explain, not guess: it tests your own explanation, not a multiple-choice click

Honest grading: marks each answer incomplete, incorrect, or uncertain

Grounded in your sources: the grade is checked against your own material, not a general answer key

Memory that persists: unresolved mix-ups carry into the next session instead of a fresh start

Grouped assessments: bundles 2 to 4 questions and scores each one separately

How DeepTutor grades you

DeepTutor is a self-hosted tutoring app (Python backend, Next.js 16 frontend) built around one loop: you read or discuss your source material, then explain it back, and the tool grades that explanation against the same material instead of a fixed answer key.

Grading comes back as incomplete, incorrect, or uncertain, and whatever you got wrong stays attached to your profile. The next session opens with those unresolved points instead of starting the subject over. The latest release (v1.6.15) adds grouped assessments that bundle 2 to 4 questions and score each one on its own, plus bounded tutoring memory so history doesn't grow forever.

The method is written up in the project's own arXiv paper (2604.26962). The README doesn't publish independent benchmark numbers for grading accuracy, only its own release notes.

The catch: DeepTutor doesn't ship its own model. You pick a provider from its built-in catalog and connect an API key, or use a local model where the provider list supports one. The tool itself is free and Apache 2.0 licensed, but the actual tutoring intelligence still has to be paid for or self-hosted by you.

Requires: Python 3.11+, Node for the Next.js 16 frontend, and a model provider API key (or a supported local model). This README doesn't list GPU, OS, or memory requirements.

Use it instead of flashcard decks when you want graded, sourced explanations rather than flat recall drills. Don't use it if you want a fully offline tutor with zero setup: you still need to configure a provider first.

Try it:
git clone https://github.com/HKUDS/DeepTutor
then follow the Get Started guide in the repo to add a provider.

https://github.com/HKUDS/DeepTutor

DeepTutor vs NotebookLM (Gemini Notebook)

Both turn your own documents into an AI study companion that answers and quizzes you strictly from your sources. People weigh them against each other because DeepTutor adds explanation-grading and cross-session memory where NotebookLM offers polished audio/video overviews and mind maps.

Pick DeepTutor if you want self-hosting, provider choice, and graded explanations that track your mistakes over time

Pick NotebookLM (Gemini Notebook) if you want a zero-setup Google tool for fast audio overviews, mind maps, and slide decks from a pile of sources

Price and hosting
DeepTutor: free, Apache-2.0, self-hosted
NotebookLM (Gemini Notebook): free tier; Plus $4.99-$199.99/mo

Model provider choice
DeepTutor: any provider, local models too
NotebookLM (Gemini Notebook): Gemini models only, no choice

Core strength
DeepTutor: grades your own explanations
NotebookLM (Gemini Notebook): audio/video overviews, mind maps

Who built it, setup
DeepTutor: HKU lab, pip/Docker install
NotebookLM (Gemini Notebook): Google Labs, sign in only

Side by side, in full

Price and hosting
DeepTutor is Apache-2.0 licensed and free to self-host via pip, Docker, or source, so the only cost is whatever model provider or local model you bring.
NotebookLM (Gemini Notebook): NotebookLM has a free tier (100 notebooks, 50 sources each) plus paid Google AI tiers at $4.99, $19.99, and $99.99 to $199.99 a month that raise notebook and source caps.

Model provider choice
DeepTutor lets you pick a provider from its own catalog, including OpenAI, Anthropic, Google Gemini, DashScope, and local runtimes via Ollama, LM Studio, or llama.cpp.
NotebookLM (Gemini Notebook): NotebookLM runs exclusively on Google's own Gemini 3 model across every paid tier, with no option to swap in another vendor's model.

Core strength
DeepTutor's grouped assessments present two to four questions at once, grade each explanation as incomplete, incorrect, or uncertain against your own source material, and carry unresolved mix-ups into the next session instead of restarting.
NotebookLM (Gemini Notebook): NotebookLM's signature strength is turning a pile of up to 50 to 600 sources (by tier) into podcast-style audio overviews, video overviews, slide decks, and mind maps with citations.

Who built it, setup
DeepTutor comes from HKUDS, a university research lab, and requires installing and configuring it yourself plus supplying your own API keys or local models.
NotebookLM (Gemini Notebook): NotebookLM is a Google Labs product you just sign into with a Google account, with zero install and all compute handled by Google.

DeepTutor: https://github.com/HKUDS/DeepTutor
NotebookLM (Gemini Notebook): https://notebooklm.google.com/

Your coding agent can paint now, and animate

Watch it turn one art style into real motion: Van Gogh's sky spinning, mosaic tiles flowing across stone, even its own creator turned into a pixel hero.

▶️ Watch the video on YouTube

Paint by code: feed it an art style, it writes the motion code

Moving masterpieces: Van Gogh's sky spins, mosaic tiles flow across stone

Its creator, pixelated: a full Super Mario level, power-ups and all

36 styles, 9 narration grammars: Mario is just one demo among many

A gate against slideshows: ships only if the motion is real, not just cuts

How it works

huashu-art-motion is a skill for a coding agent (built around Claude Code), installed with npx skills add alchaincyf/huashu-art-motion. Tell it a style, it designs one frame, then writes scene-by-scene motion code and renders it with Playwright's Chromium.

The skill itself is free and MIT-licensed. The real cost is the coding agent that drives it: a Claude Code subscription or API key (in the video, even grabbing a power-up means picking Claude over OpenAI first), or whatever agent you connect.

By the project's own README: 53 style recipe cards (covering dozens of styles, several anchored to real museum artworks), 53 matching scene scripts, 9 narration-grammar cards, and 17 animation library modules. Its own qa.py script checks stability, motion and clarity before delivery, and a separate film-gate script rejects a cut if it reads as slideshow motion instead of real movement, by the authors' own description.

Requires: a coding agent, plus uv, ffmpeg, and Playwright's Chromium (install with uv run --with playwright playwright install chromium). The README lists no GPU or OS requirement.

Catch: this is an engine for an agent to use, not a standalone app. Scenes with a human figure still need frames from a separate image-generation model; the code only animates them.

Try a single frame:

uv run --with playwright python scripts/engine/render.py --solo 09_postimp --stills 0.3 --out test

Use it instead of hand-animating art references in After Effects or Remotion when you want code-driven, reproducible motion. Don't use it if you just want a quick one-off clip without installing an agent and a dev toolchain.

License is MIT for the code, with named exceptions: the fonts (OFL), and the author's own character frames and the Clawd fan animation, none of which ship under MIT.

github.com/alchaincyf/huashu-art-motion

⚡ AI News
Anthropic reports early wins in AI drug discovery: Claude cut one CAR-T gene design by 70% and sped up Enveda's regulatory paperwork by 89%, Anthropic says.

Anthropic's S-1 reveals outsized founder, trust control: A Long-Term Benefit Trust and the founders keep majority board power, leaving public shareholders little say.

Google DeepMind launches Genie 4 world model: Genie 4 adds real-time audio and free-form editing, and holds a consistent scene for two full minutes.

A $300/month AI team, running free on your desktop

OpenBot runs the same kind of persistent AI agent team you'd get from xAI's Grok Bot, but entirely on your own machine.

▶️ Watch the video on YouTube

A team, not a chatbot: each agent keeps its own name, instructions and memory

Agents talk to each other: they hand off tasks, share files, and one can drive a built-in browser

Bring your own brain: plug in Claude Code, Codex, Grok CLI, Gemini, or a fully local model

No OpenBot subscription: the app is free, you only pay for whichever AI provider you connect

Wallet makers, not an AI lab: its creators built Nightly, a multichain crypto wallet, first

How OpenBot works, and the catch

OpenBot is a desktop app (Electron, SolidJS, Bun, SQLite) for macOS, Windows and Linux, plus a Docker image. It spins up a small team of persistent agents, each with its own name, instructions, working directory and conversation history. They message each other, hand off tasks and share files, and one of them can drive a built-in browser.

It ships no model of its own. You connect Claude Code, Codex, Grok CLI, Gemini, OpenCode, Cursor CLI, Cline CLI, any OpenAI-compatible endpoint, or a fully local model through Ollama or LM Studio. Workspaces, chats and browser data stay on your machine either way.

The team's own Product Hunt pitch frames OpenBot as a free, local alternative to xAI's Grok Bot, which runs $200/month inside Cursor Ultra or $300/month on SuperGrok Heavy, with extra usage billed on top and no spend cap, according to a third-party pricing breakdown. That comparison is the maker's framing, not an independent benchmark.

Requires: an API key or subscription for whichever provider you connect, nothing extra if you run a local model via Ollama or LM Studio. The README names no GPU, RAM or chip requirement beyond the three desktop platforms plus Docker.

The catch: starting with newer releases, the code moved from Apache-2.0 to PolyForm Noncommercial 1.0.0, so commercial use needs a paid license from Synthetify Labs. Only v0.1.11 and earlier stay fully open source. On GitHub it currently sits around 2.4k stars, 245 forks and 187 open issues.

Use it instead of Grok Bot when you want the same multi-agent setup without a metered cloud bill. Don't use it if you need a commercial deployment under the current license, or want a vendor-supported cloud product.

To try it, clone the repo and check the Releases page for a packaged build per platform:

git clone https://github.com/nightly-labs/openbot.git

github.com/nightly-labs/openbot

⚡ AI News
Anthropic deepens Claude discount deal for US government: Anthropic now gives every US federal, state and local agency a 40% Claude discount, citing $1B in savings.

Parents sue OpenAI over ChatGPT medication death: Parents allege ChatGPT coached their epileptic daughter to taper off her seizure medication before she died.

Opera proposes WebAI standard for local AI in browsers: Opera's new WebAI standard lets browsers run open models like Gemma locally, without needing a cloud server.

An open-source InDesign alternative, built in pure Rust

DesignCraft is a full desktop publishing and layout editor, written entirely in Rust.
It threads text through columns, wraps it around pull quotes, and can hand the whole job to an agent.

▶️ Watch the video on YouTube

Own it outright: open-source layout editor, built in pure Rust

Familiar workspace: the panels, tools and shortcuts of real publishing software

Threaded text: flows through columns and wraps around pull quotes

Line breaks that hold: its paragraph composer keeps them identical on screen and in PDF

Claude can design pages: every menu runs over its own MCP server

What's inside

A Knuth-Plass paragraph composer, dictionary hyphenation (the public domain Moby word list plus the project's own trained patterns), optical margin alignment, columns, baseline grid, tabs, rules and shading. Line breaks match exactly between the on-screen view and the exported PDF.

One Rust codebase runs the desktop app on macOS, Windows and Linux, and compiles to WebAssembly for the browser. Every menu item, tool gesture and dialog can also be driven over a JSON control channel and an MCP server, so Claude or another agent can lay out pages itself.

The editor is free and open-source (MIT OR Apache-2.0): no subscription, no licence server, no telemetry. Handing the layout job to an agent needs your own Claude access or another MCP-capable model. That part isn't free.

Requires: Rust 1.95 or newer to build the desktop app (the egui frontend), plus a Rust toolchain and cargo. The project doesn't list extra hardware requirements.

The catch: this is a brand-new project, already carrying 724 open issues on its tracker.

Try it: cargo run --release -p designcraft -- --sample opens the sample magazine; cargo run --release -p designcraft-cli -- run --sample --all-pages out/ renders every page headless to PNG.

Use it instead of Scribus or InDesign if you want a scriptable, agent-drivable layout tool in Rust. Don't use it yet if you need a mature, issue-free production pipeline.

https://github.com/storytold/designcraft

You rank on Google. AI search might skip you anyway.

Claude SEO is a free Claude Code plugin that checks whether both Google and AI search find and trust your site, then hands you a ranked plan.

▶️ Watch the video on YouTube

Full SEO sweep: crawl errors, Core Web Vitals, schema, AI citations

Parallel agents: one command, up to seventeen agents audit your site at once

One ranked plan: results in ten to fifteen minutes, sorted by impact

Its own proof: claude-seo.md got the same audit from day one: 10.8K clicks, a 99 score, in twelve weeks

The catch: needs Python 3.11 through 3.14, plus its own Chromium

Inside the audit

Claude SEO is a Claude Code plugin built from 26 sub-skills and 19 specialist agents (Sonnet and Opus). It covers technical SEO, E-E-A-T, schema, GEO/AEO (AI Overviews, ChatGPT, Perplexity, Gemini), agent readiness, backlinks, local and e-commerce SEO.

Run /seo audit <url> and up to 17 agents work the site at once. 10 to 15 minutes later you get one prioritized action plan; each item explains why it matters.

The 10.8K clicks and 99 mobile PageSpeed score are the project's own figures, from running the tool on its own site, claude-seo.md, since launch, over its first 12 weeks. That is the authors' own result, not an independent benchmark.

Requires: Python 3.11 through 3.14 and its own Chromium runtime (set up once via /seo setup), plus Claude Code itself. The plugin is free; Claude Code needs a Claude subscription or API key, so the model calls behind each audit are not free.

The catch: this is a local Python runtime plugin, not a hosted service, so the one-time setup (Python version, Chromium download) has to succeed before the first audit. Nine optional extensions add live SEO data but need their own API keys.

Use it instead of a manual SEO consultant pass when you want a scored, ranked fix list in minutes. Skip it if you don't already run Claude Code, or if you need results without any setup step.

Try it: /seo setup, then /seo doctor to check the runtime, then /seo audit https://yoursite.com.

AgriciDaniel/claude-seo

⚡ AI News
Anthropic launches Smart Reports for AI spend: Anthropic's new Smart Reports track Claude ROI for teams, as the company posts 900% net revenue retention and a Q2 operating profit.

Google opens Gemini and Search to outside apps: Google's new Connected Apps Platform lets developers plug services into Gemini and Search, building on 50+ existing partners.

Nvidia releases open Nemotron 3.5 Transcribe model: Nvidia's open-weight Nemotron 3.5 Transcribe handles 40 languages and an hour of audio, topping the Open ASR leaderboard.

Cursor just turned its AI agent into an app store

Your coding agent couldn't talk to Stripe or query Datadog logs. Cursor shipped something called Plugins, right inside the editor, to fix exactly that.

▶️ Watch the video on YouTube

Rules, skills and MCP in one install: a plugin bundles everything the agent needs to use an outside tool

100+ plugins already live: Stripe, Figma, Datadog, GitLab, installed from the Marketplace or /add-plugin

Not the old VS Code marketplace: classic editor extensions still come from the separate Open VSX registry

Built together with rivals: AWS, Microsoft, OpenAI and Vercel shipped one open plugin format

One catch: plugins only run from Cursor 2.5 onward

What a plugin bundles

A Cursor plugin packages rules, skills, subagents, MCP servers, hooks and commands together, so the agent gets both instructions and a live connector to act on a tool like Stripe, Linear, AWS, Slack or Figma, per Cursor's own plugin docs.

The Marketplace launched with Cursor 2.5 on Feb 17, 2026, with partner plugins from Amplitude, AWS, Figma, Linear, Stripe, Cloudflare, Vercel, Databricks, Snowflake and Hex, according to Cursor's launch blog post. A later wave added 30+ more, including Datadog (query logs, metrics and traces in plain language), GitLab, Atlassian and Glean. Cursor's own cursor/plugins GitHub repo, which hosts the spec, has passed 14,500 stars and lists 100+ plugins, per the repo itself.

In August 2026, Cursor joined a five-company steering committee with AWS, Microsoft, OpenAI and Vercel to ship "Agent Plugins 1.0", an open, vendor-neutral format. Build a plugin once and, per AWS's own announcement, it can run in Cursor, VS Code, Kiro or any other client that implements the same spec.

Cost: installing a plugin itself is free. Cursor is a separate paid product with a free tier. The README doesn't say whether Plugins need a paid plan. Each connected service, Stripe, Datadog, AWS, still needs its own account and API key.

Requires: Cursor 2.5 or later; an account and API key for whatever service a plugin talks to; the project doesn't list any extra hardware requirements.

The catch: plugins are a separate layer from the classic VS Code-style extensions Cursor already installs. Those still come from Open VSX, not Microsoft's marketplace, and nothing here works before Cursor 2.5.

Use it instead of hand-writing MCP configs for every tool, when you want the agent wired into your stack in one install. Don't use it if you're on an older Cursor build, or if you only want classic editor extensions.

Try it inside Cursor with /add-plugin, or browse plugins at cursor.com/marketplace.

https://github.com/cursor/plugins

⚡ AI News
Microsoft discloses $10B quarterly revenue from OpenAI: Microsoft earned $10B from its OpenAI deal last quarter, nearly 11% of its total revenue, with $7.2B still owed.

Google's Gemini app hits 1 billion monthly users: Alphabet's Q3 revenue rose 25% to $127.6B, with Cloud up 103% and Gemini now at 1 billion monthly users.

Anthropic expands AI for Science to 20 institutions: Anthropic grew its free Claude credits program to 20 science funders, research institutes and projects like FlyWire and gnomAD.

One repo, dozens of apps: open-source clones of the SaaS you pay for

The video opened on a wall of app screenshots: this project rebuilds the office and collaboration tools you already pay for, from scratch, as open source.

▶️ Watch the video on YouTube

Office, rebuilt: Spreadsheet, Documents and Presentation stand in for Excel, Word and PowerPoint

Meetings and chat, too: Meetings, Teamchat and Email take on Zoom, Slack and Gmail

Reaches further: README also lists Photoshop and AutoCAD style apps in Creative Tools, Engineering

The catch: a few apps, like Notes and Signatures, ship source only, with no live demo yet

Already live: Office, Meetings and Teamchat apps are free, open source and usable today

What's in the repo

Everything is a single monorepo where every app lives in its own folder under src/: Spreadsheet, Documents, Presentation, Whiteboard, PDF, Notes and Records cover office work; Meetings, Teamchat, Email, Contacts, Calendar and Signatures cover collaboration. The README's table of contents also names Creative Tools, Engineering, Productivity, Finance, Business, Developer Tools and Games sections, which is where the Photoshop and AutoCAD style apps from the video belong.

There's no benchmark or performance number here. This is the project's own stated goal: to rebuild and open-source the world's software, counted in apps shipped, not a measured speed or accuracy claim by anyone.

These are plain self-hosted web apps, not AI wrappers, so the tool itself is free with no model or API bill attached. Some apps already have a hosted demo you can open straight in a browser, for example spreadsheet.puter.com and meetings.puter.com. Others, like Notes and Signatures, ship source only, with no live demo link yet.

Requires: Node.js 22.12 or newer, git, npm.

git clone https://github.com/HeyPuter/everything.git
cd everything/src/[app-name]
npm install
npm run dev

Use it instead of Microsoft 365, Zoom or Slack if you want a free, self-hosted stand-in you can run yourself. Don't expect it to replace them today if you need their polish, integrations or mobile apps.

github.com/HeyPuter/everything

The Everything Project vs Nextcloud Hub

Both pitch self-hosted, open-source replacements for your SaaS stack: Puter's Everything project rebuilds Office, Zoom, Slack, Photoshop and AutoCAD from scratch as separate MIT apps, while Nextcloud Hub bundles office, chat, file sync and calls into one mature platform used by millions. Anyone picking a self-hosted Microsoft 365 or Google Workspace alternative weighs these two.

Pick The Everything Project if you want free MIT source for dozens of app categories, including creative and CAD tools, to adapt piece by piece, even if some are demo-less and early.

Pick Nextcloud Hub if you need a production-ready, integrated office, chat and video suite today, with enterprise support and a vendor used by governments at scale.

Price and licensing
The Everything Project: free, MIT, no paid tier
Nextcloud Hub: free self-host; from €71/user/yr

Setup and hosting
The Everything Project: clone + npm per app
Nextcloud Hub: one Docker command

Breadth vs maturity
The Everything Project: most categories, many demo-less
Nextcloud Hub: fewer apps, production-proven

Who's behind it
The Everything Project: Canadian startup, ~7 staff
Nextcloud Hub: German firm, since 2016

Side by side, in full

Price and licensing
The Everything Project: Every Everything app is released free under the MIT license with no subscription, hosted paywall, or enterprise tier offered by HeyPuter.
Nextcloud Hub: Nextcloud's self-hosted Community edition is free with full features, but its Enterprise support subscriptions start at €71.29 per user per year and require a minimum of 100 users.

Setup and hosting
The Everything Project: Each app lives in its own folder; running one locally means Node.js 22.12+, then git clone, npm install and npm run dev per app, with no shared backend linking them together.
Nextcloud Hub: Nextcloud All-in-One starts the whole office-plus-chat server with a single docker run command, though Talk's video calls recommend adding a TURN server for reliable connections.

Breadth vs maturity
The Everything Project: The README lists nine categories reaching into Photoshop, AutoCAD, SolidWorks and music-production alternatives that Nextcloud has nothing like, but Photo Studio, Video Studio, Vector Studio and 3D Studio all still show source code only with no live demo.
Nextcloud Hub has no creative or CAD apps, but its office, chat and file-sync core is a decade old and runs on over 400,000 servers, including rollouts to hundreds of thousands of French and German government staff.

Who's behind it
The Everything project comes from Puter Technologies, a Canada-based startup founded in 2022 with roughly seven employees and one undisclosed pre-seed round from Accel Starters, Shilling Capital and others.
Nextcloud Hub: Nextcloud GmbH is a German company headquartered in Stuttgart, founded in 2016 by Frank Karlitschek (who earlier founded ownCloud), and now markets itself as the sovereign European alternative to Microsoft 365.

The Everything Project: https://github.com/HeyPuter/everything
Nextcloud Hub: https://nextcloud.com

A remote control for your coding agents

You start a coding agent, then end up babysitting it. T3 Code puts a remote control for it on your phone.

▶️ Watch the video on YouTube

No AI inside: it drives the coding agents already on your machine, it doesn't think for itself

Six agents, your subscriptions: Claude Code, Codex, Cursor, Grok Build, OpenCode, Google Antigravity

Phone, web, or desktop: kick off a task at your desk, then steer it from anywhere

Early days: expect bugs, and an agent needs to already be installed and logged in

One control surface for agents you already run

T3 Code runs no model of its own. It's a harness that connects to Claude Code, Codex, Cursor, Grok Build, OpenCode, or Google Antigravity on your computer, and lets you drive whichever is set up from an iOS or Android app, a web app, or an Electron desktop app.

The tool itself is free and open source. The intelligence isn't: each agent still runs on your own subscription or API key for Claude, OpenAI, Cursor, xAI, or Google, whatever you already use.

Requires at least one agent already installed and authenticated before you open T3 Code: for example Claude Code with claude auth login, Codex CLI with codex login, or OpenCode with opencode auth login. Antigravity just needs enabling in Settings and a Google sign-in, no CLI. The project doesn't list OS, chip, or memory requirements beyond that.

The catch: the authors call the project "very very early" and say to expect bugs. It also isn't accepting outside contributions yet beyond small fixes.

Try it with curl -fsSL https://t3.codes/install.sh | sh (PowerShell on Windows: irm https://t3.codes/install.ps1 | iex), then run t3. To test it without installing, run npx t3@latest. Desktop builds are also on GitHub Releases, winget, scoop, Homebrew, and the AUR.

Use it instead of leaving a terminal tab open all day when you want to check on a task from your phone. Don't use it if none of those six agents are set up yet, since T3 Code has nothing of its own to run.

https://github.com/pingdotgg/t3code

⚡ AI News
Mistral releases open weights for 1T-parameter Large 4: Mistral ships full weights for its trillion-parameter Large 4 model, tuned for cybersecurity, coding and finance.

Deepgram acquires Daily, maker of Pipecat voice framework: Deepgram buys Daily, the startup behind the open-source Pipecat voice AI framework, with deal terms undisclosed.

Wabi hits $1.25B valuation building mini-app AI agent: Wabi, the mini-app-building AI agent from Replika founder Eugenia Kuyda, raised $55M at a $1.25B valuation.

Macro: the PDF editor that open-sourced its whole work OS

You're paying for Slack, Notion, Linear and Salesforce, all at once.
Macro wants to replace all of them, in one open-source app.

▶️ Watch the video on YouTube

27+ apps, one app: Slack, Notion, Linear and a CRM bundled into Macro

From PDF editor to work OS: started in 2020, now email, chat, tasks, calls, CRM

Shared team memory: one AI agent acts across every module

$30M+ from a16z, then open-sourced: the whole app, AGPLv3, not open core

$200/month plan, free to self-host: model costs aside

What self-hosting means

Macro started in 2020 as an AI editor for PDFs and DOCX files that flagged key terms, cross-references and errors. In 2026 it pivoted into a full work platform: email, chat, docs, tasks, calls and a CRM, tied together by "shared team memory" so one AI agent can act across every module.

The company raised a $9.3M seed in 2023 and a $12M Series A in 2024, both led by Andreessen Horowitz, $30M+ total (TechCrunch, Law.com). It then released the whole platform under the AGPLv3 license, which the project's own GitHub page calls "not open core": the same code the $200/seat/month Max plan runs on, not a stripped-down free tier.

The GitHub repo has reached 5.6k stars and 533 forks. The frontend is SolidJS (web, Tauri desktop, mobile); the backend is Rust, split into 42 separate deployable services, and that split is the catch: this is not a one-container app to self-host.

Requires: the project doesn't list hardware requirements beyond a server able to run those 42 Rust services, plus either a local model or a paid API key, since self-hosting the app does not make the AI itself free. For comparison, Macro's own cloud tiers: Free (Gemini Flash only, 5GB storage, 2 email accounts), Pro at $40/seat/month, Max at $200/seat/month.

Use it instead of a Slack plus Notion plus Linear plus CRM stack when you want one app and your own data under your own roof. Don't use it yet if you need a proven, simple hosting setup: forty-two services is a lot to run solo, and the AI itself is never free, only the app is.

Clone the repo and follow its setup docs to self-host:
git clone https://github.com/macro-inc/macro

github.com/macro-inc/macro

AI Made Developers 19% Slower, But They Felt 20% Faster
A walk through the trial, the rerun, the vendor benchmark and the paper behind the 19% argument.
Our Short on this: Macro: the PDF-editor startup that open-sourced an entire work OS

Watch the video on YouTube
Read the article