Is Froots ready to be my daily driver?

If you write in markdown and you're comfortable using a beta — yes. The editor, vault, memory, git history, skills, and agent are all shipping. If you need multi-device sync, mobile, or collaborative editing, those are on the Summer 2026 roadmap. You can keep a Froots vault alongside Obsidian or Bear while we close those gaps — the format is plain markdown.

What do I need to start?

A Mac or Windows machine, and an Anthropic or OpenAI-compatible API key. We route to Claude by default; you can point at any OpenAI-compatible endpoint including local runtimes like Ollama, LM Studio, or llama.cpp. No account signup required.

What happens to my data?

It stays on your machine. Froots writes plain .md files into a workspace folder you choose, auto-commits every change to a local git repo, and encrypts API keys in your OS keychain. No cloud sync in 0.1 — nothing leaves your machine unless you explicitly point an integration at it.

Is there a free tier?

Yes. During beta, every feature is free — you just bring your own API key. Managed model routing, multi-device sync, and team features ship with the 1.0 launch later this year; a free-forever Seedling tier will remain.

What's the version number and how often will you ship?

This is 0.1.0. We're pacing releases by apple variety — this one's nameless; 0.2 is on deck. Point releases weekly for bug fixes, minor versions monthly for features, majors when something substantial lands. The full changelog lives on this blog.

Where do I report bugs?

For now: email hello@froots.ai with 'beta bug' in the subject, and include your OS + what you were trying to do. We're hiring a community manager specifically for beta triage; once the Discord opens, reports move there.

Froots 0.1 Beta — the first public build

Froots is in public beta. A local-first desktop workspace — markdown editor, git-backed vault, customizable agent, and 60+ built-in skills — honest about what's shipping today and what's still ahead.

We're in public beta. Froots 0.1 is shipping today to anyone who wants to run it — with the caveats that beta implies and the honesty we think they deserve.

This is not a 1.0 announcement with a preloaded marketing engine. It's a real first build of an app we've been working on for a year, enough features shipping that it earns "beta" instead of "alpha," and enough gaps that we want to tell you about them before you install.

What ships in 0.1

A local-first markdown workspace. Your notes are plain .md files in a folder you pick. Every save is auto-committed to a local git repo inside the workspace. Nothing is in a proprietary format; if Froots disappeared tomorrow, you'd still have every byte you ever wrote in a format any editor can open.

A capable block editor. Tiptap under the hood, rendered in a single unified surface with:

Wikilinks, backlinks, and an "unlinked mentions" drawer that finds pages with your current title that haven't been linked yet
Syntax-highlighted code blocks (Shiki, 150+ languages)
LaTeX math via KaTeX (inline $ and block $$)
Mermaid diagrams rendered live
Tables with keyboard navigation
Task lists with [ ], [x], [/], [?] states

An always-on agent with real memory. The right pane is an assistant that can read your vault, edit files, hit connected integrations, and talk to any Claude-compatible or OpenAI-compatible endpoint. What makes it useful beyond "a chat window in your notes app":

Three editable memory files (context.md, decisions.md, learnings.md) that the agent writes to as it learns, and reads back on every turn via local semantic search. Your vault grows a memory without you thinking about it.
Four persona files (IDENTITY.md, USER.md, SOUL.md, BOOTSTRAP.md) you can open in any editor. Want a tersely-worded, never-apologize agent? Rewrite IDENTITY.md, save, the agent updates mid-session.
60+ built-in skills — connectors for iMessage, Gmail, Slack, Discord, Telegram, Signal, WhatsApp, GitHub, Notion, Obsidian, HomeKit, Sonos, Spotify, Apple Notes, Things, 1Password, Blender, Canvas, PDF, OCR, and more — each sandboxed, each toggleable.
A skill-creator you can invoke to stub out new skills when the built-ins don't cover you.

Three trust levels. Careful prompts before every tool call. Balanced prompts once per session for destructive actions. Yolo doesn't prompt. Per-tool scoping lets you allow reads but not writes (or shell, or network) for any agent.

A unified inbox. Connect any subset of: iMessage (macOS Full Disk Access required), Gmail (OAuth), any IMAP/SMTP provider, Slack (Socket Mode, no webhook), Discord, Telegram, Signal (via signal-cli), WhatsApp, X DMs. Triage by keyboard, summarize by thread, let the agent draft replies for you.

Native OS widgets. The agent can emit a widget on demand — SwiftUI WidgetKit on macOS, Adaptive Cards on Windows. "A widget showing my next meeting" returns a native tile you can pin to Notification Center or the Windows Widget Board.

Global hotkey. ⌘⇧Space (Mac) / Ctrl⇧Space (Windows/Linux) opens Froots from anywhere. Rebindable.

What doesn't ship yet

Beta honesty: if one of these is a hard requirement for you, wait for 1.0.

Mobile. iOS and Android apps are in development. No public build yet.
Sync. No multi-device or cloud sync in 0.1. Your vault is local-first; if you want it elsewhere, point the workspace folder at iCloud Drive / Dropbox / a git remote. Native CRDT sync with end-to-end encryption is on the 1.0 roadmap.
Teams. Single-user only today. Shared vaults, cross-seat billing, managed cloud hosting — later this year.
Routine canvas. Highlight-an-English-sentence-and-get-a-runnable-flow-graph is in internal testing. You can already schedule tasks, chain skills, and compose prompts — the visual canvas just isn't the entry point yet.
Public plugin SDK. The internal skill system is stable; the public TypeScript SDK for third-party skills is hardening before release.
Notifications. In-app and native notifications for memory saves, KB edits, permission requests, and scheduled task completions are planned and partially wired — not on by default.

None of these are vaporware. They're all designed; most are partially implemented. We'd rather ship 0.1 now and close these gaps in the open than keep it in private beta for another six months.

Under the hood

A few honest architecture notes, since beta users tend to care:

Desktop shell: Tauri 2. Real native windows, real IPC, not an Electron browser. The binary is ~50 MB; RAM baseline is ~180 MB.
Memory index: the agent's memory (and your vault) are chunked by markdown heading, locally embedded with a bundled sentence model, and stored in a local SQLite-family vector database. Top-k relevance is re-scored on every turn — injection takes <50 ms on a modern laptop.
Model layer: we talk to your chosen model over its standard API. No proxy, no routing layer in between, no rate limits on our side. Your keys live in your OS keychain.
Storage: nothing proprietary. Open the workspace folder in any editor. Every file is auditable. Every change is in a local git log.
Sandbox: skills run in a subprocess with declared filesystem and network capabilities, not in the main app process. A crashed or malicious skill can't take the app down.

How to get it

Visit the downloads page and click your platform. For the duration of the beta, installers are distributed via the waitlist — drop your email and we'll ship a signed build within a business day, along with a beta Discord invite.

The installers are free during beta. Pricing kicks in at 1.0.

The shape of the next few releases

0.2 (late May 2026) — inbox UI polish, notification system goes from partial to shipped, first-run onboarding pass, memory panel UI.
0.3 (June 2026) — iOS beta (TestFlight), routine canvas preview, first community skill registry.
0.4 (July 2026) — Android beta, sync preview on Froots Cloud managed tier.
1.0 (Autumn 2026) — full sync with E2E encryption, teams tier, public plugin SDK, a name that isn't 0.x.

Why we're doing this

Because the note-taking apps we love don't have agents, and the agent apps we love don't have editors. Because markdown is the right format and nobody should take it from you. Because a workspace that keeps memory across sessions starts to feel like a second brain about six weeks in, and most of us have been waiting for that for a decade.

0.1 is where we start. Thank you for trying it.

— The Froots team