NabuSuite

Native, local-first apps for people who build. NabuCode · NabuVoice · NabuSound · NabuStats. Lifetime licenses, no cloud. Your data stays on your machine.

[NabuSuite @NabuSuite]

We're NabuSuite. Native, local-first apps for people who build.

NabuCode: agents keep running when you close the app
NabuVoice: dictation that never leaves your Mac
NabuSound: every app gets its own volume
NabuStats: a monitor that thinks in apps

Your data stays local.

[NabuSuite @NabuSuite]

@viticci @bot @cursor_ai @macstoriesnet The self-testing part is what I'd like to hear more about. When a PR needs a judgment call, do you read the diff on the phone, or trust the test run and merge? Two months of driving it from a phone is a real data point on how much review can actually move off the desk.

[NabuSuite @NabuSuite]

@LKang37857 Nice, small and focused. Does it take epoch seconds and milliseconds too, or ISO strings only? Half the timestamps I paste come out of logs as 1759... and I never remember which unit they are. The menu-bar version sounds like the right home for it.

[NabuSuite @NabuSuite]

@trq212 The blocked-while-offline case cuts both ways. Today the agent is blocked the moment the laptop closes; in the cloud model it's blocked when the laptop is away. Either way something has to own the in-flight session while one side is gone. Curious which side you'd put that on.

[NabuSuite @NabuSuite]

NabuVoice is a one-time lifetime license, 3 Macs, $19.99 at the launch price. Text lands in whatever app has focus: https://t.co/1X259TZXKR

[NabuSuite @NabuSuite]

Dictation, offline.

NabuVoice transcribes with one of 13 speech models running on your Mac. The cleanup pass is a local LLM through llama.cpp on Metal. Once a model is downloaded, transcription needs no network. Your audio never leaves the machine.

Apple silicon, macOS 12+.

[NabuSuite @NabuSuite]

@timniverse Agreed. One framing that helps: ask for "things a reviewer should check by hand" instead of "what you didn't verify". The first reads as being helpful, the second as admitting a miss, and the model fills in the helpful one more readily. Still worth spot-checking the diff.

[NabuSuite @NabuSuite]

@thdxr One thing that drifts less than a doc: the rejected options, as a comment on the line that replaced them. "Tried 3, broke resume after sleep." An agent reads it right before editing, and it dies with the code. Doesn't cover decisions that span files. What do you do there?

[NabuSuite @NabuSuite]

NabuStats lives in the menu bar or tray, with history you can scroll back through. One lifetime license covers 3 computers, $9.99 at the launch price: https://t.co/XOeIlAgxVX

[NabuSuite @NabuSuite]

Ten node processes. Which project do they belong to?

NabuStats rolls dev servers, test runners and coding agents up under the project folder they run in. Docker containers sit beside your apps, each with its own CPU and memory.

macOS, Windows and Linux.

[NabuSuite @NabuSuite]

@reach_vb Clear primer, thanks. One question: does the reviewing agent see the whole task so far, or only the single action it is judging? That context seems like the difference between a safe command and a safe command for this job.

[NabuSuite @NabuSuite]

@dhh @Lufthansa_USA The split makes sense: the laptop is a screen, the shed does the work. The part worth getting right is that sessions there outlive the connection, so a hotel wifi drop doesn't take an agent down mid-task. How's the latency from the road?

[NabuSuite @NabuSuite]

NabuSound runs on macOS 14.2 or later, Apple silicon and Intel. One lifetime license covers 3 Macs, $9.99 at the launch price: https://t.co/EDbHBN7OPC

[NabuSuite @NabuSuite]

A tip for calls: give each app its own output.

In NabuSound, send Zoom to your AirPods and Spotify to the speakers. The music dips on its own while you're on a call and comes back when you hang up.

If a device disconnects, that app falls back to the system output.

[NabuSuite @NabuSuite]

@theo @davis7 The theme tokens are the smart part: a generated chart that matches the app stops looking like a bolt-on. Is the agent's output rendered in a sandboxed frame with the tokens passed in, or inline in the thread?

[NabuSuite @NabuSuite]

NabuCode runs Claude Code, Codex, OpenCode and Grok CLI in one tiling grid on your computer; the phone app supervises. Pro is $9.99/mo at the launch price, locked for life if you subscribe before Nov 23: https://t.co/IS4BCgHPOb

[NabuSuite @NabuSuite]

An agent waiting on a yes-or-no is an agent doing nothing.

The NabuCode phone app lists every running session with its status and live output. When one stops to ask, answer from your phone and it carries on.

Paired by QR code, over TLS. iOS and Android.

[NabuSuite @NabuSuite]

@ThePrimeagen One thing that helps with the slow half: let the agent find the path once, save the inputs plus a screenshot check per step, then replay it without the model. Call the agent again only when a check fails. You pay the 2 minutes when the UI changes, not on every run.

[NabuSuite @NabuSuite]

@ggerganov Congrats on the release. Is the Metal speedup across the board, or does it mostly show on larger models? We run a 1.5B model by default for on-device text cleanup, so small is where we'd feel it most.

[NabuSuite @NabuSuite]

NabuVoice comes with 13 local speech models and optional cleanup by a local LLM. One lifetime license covers 3 Macs, $19.99 at the launch price: https://t.co/1X259U0vAp

[NabuSuite @NabuSuite]

A dictation tip: teach it your words before you start.

In NabuVoice, add the names, jargon and acronyms you use to Vocabulary. Then make a Snippet, and a short spoken phrase expands into a full block of text.

Transcription runs on your Mac. Apple silicon only for now.