Product Features Compare with LM Studio and Ollama Pricing Changelog Guides Which model fits your Mac Local AI on an 8 GB Mac What local AI costs More Source on GitHub Contact

A council of models.
One answer.

Several models running on your own Mac argue one question out, and one of them writes the answer. Offline by default.

Free, MIT-licensed core · macOS 26 on Apple silicon · signed and notarized by Apple

It all happens on your Mac.

Chat, a coding agent, images and dictation.
No account, no cloud, no telemetry.

Slate — Roundtable

A Roundtable: two seats argue, the chair writes the answer.

Made for people who already run their own models.

llama.cpp GGUF Hugging Face MCP tools Apple silicon Qwen3 Llama Mistral Gemma
The wedge

One model is a guess. A room is an answer.

Give a hard question to several models with different roles, run a few rounds of disagreement, then hand a chair model the transcript to weigh the arguments and write the reply. It is the one thing a single model, and the tools that only run one, cannot do.

  • Pick the seats, assign roles, set the rounds
  • Local seats by default, add a cloud seat only if you want one
  • Watch the debate, keep the synthesis
How the Roundtable works
live demo · scripted

"Design a rate limiter for our API."

Three seats debate, the chair synthesizes

Press run to watch a council reach one answer.
One workspace

Seven surfaces, one native app.

Each one otherwise means a separate tab or a separate purchase. Slate keeps them side by side, and keeps the work on your machine.

the wedge · free & Pro

Roundtable

Seat several local models, give them roles, and let them argue across a few rounds. Then one chair model reads the whole transcript and writes the answer.

  • Pick the seats & roles
  • Local by default
  • Keep the synthesis
everyday · free

Chat

Open a focused conversation with any GGUF you bring, drop in a file or an image, and let it read your own documents on device before it answers.

  • Any local model
  • On-device file chat
  • Editable memory
agentic · Pro

Code

Point Slate at a project. Ask reads only, Edits proposes diffs you approve, Auto runs the loop, and a checkpoint rolls you back when the plan beat the patch.

  • Ask / Edits / Auto
  • Checkpoints & Git
  • Per-project memory
generate · Pro

Image

Run local diffusion models with the prompt kept on the machine. Start from text, rework a source frame, hold a seed to iterate, then drag the result onward.

  • Text & image to image
  • Seed control
  • Fully on device
hands-free · Pro

Voice

On-device speech in and out, with turn detection and interruption handling, so it feels like a conversation instead of a dictation box. Nothing hits a cloud transcriber.

  • Local STT & TTS
  • Turn detection
  • No cloud
system-wide · Pro

Flow

Dictate into whatever app already has focus. Slate transcribes locally, cleans the result, and drops it at your cursor, in German or English, with your own vocabulary.

  • Push-to-talk anywhere
  • Local cleanup
  • DE / EN + vocab
iphone companion · coming soon

Slate Remote

Pair your iPhone with a QR code and chat with your Mac's models from the couch. Every token streams live over an encrypted link — home Wi-Fi only, nothing leaves your network. The iPhone app is not released yet; the Mac side ships today.

  • Pair by QR
  • Live streaming
  • Home network only
The privacy path

Local by default. Open by design.

"Your work stays on your Mac" is only worth saying if you can check it. So the engine is open: read it, build it, and confirm there is no hidden call home.

Read the engine on GitHub

No account

Nothing to sign up for. Open the app and work.

No telemetry

No analytics, no crash pings, no conversation ever sent to us.

Read the source

The engine is on GitHub under MIT. Verify the claim yourself.

Cloud by choice

Off by default. Turn on one connector with your own key when you want it.

Open core

Free to run. Worth paying for.

The everyday workspace is open source and free. One payment unlocks the rest, forever. No subscription, no account. What running it costs you in electricity.

Open source
€0free & MIT

The local workspace, in the open. Bring a model and start; no key, no card.

  • Local chat with any GGUF you bring
  • Chat with your files, on device
  • Roundtable with two seats
  • Code in Ask mode, read-only
Get it on GitHub
Slate ProPay once
€39.99once · every update included

Every surface unlocked, on up to three Macs. One payment, yours to keep.

  • The full Roundtable, deeper synthesis
  • Flow dictation, system-wide
  • Agentic coding, images and voice
  • Memory, model compare, local tools & MCP
Get Slate Pro
No account to buy No subscription 14-day money-back Instant licence key
After you download

Three minutes from download to first answer.

Open it

No account, no licence key, no sign-in. It is signed and notarized by Apple, so it opens on a double-click.

Take the model it suggests

Slate bundles no models, so it reads your Mac's memory and recommends one that fits: about 7 GB on a 16 GB Mac, 10 GB on 24 GB, 16 GB on 32 GB. One button, downloading in the background. Or pick your own from Hugging Face. See what fits your Mac, or the honest version for an 8 GB machine.

Ask it something

The answer is computed on your Mac. Nothing is sent anywhere unless you add a cloud key yourself, and Silent Mode blocks even that.

Needs macOS 26 on Apple silicon. On an Intel Mac or an older macOS it will not run, and we would rather you knew that now than after the download.

Bring your models. Keep your data.

Read the source today. Buy Pro once when you are ready. Everything runs on your Mac either way.