Arla

One assistant, one conversation, on your computer and your phone.

Arla remembers you, plans with you, and actually does the work, using the AI models you already have. Your data stays on your own devices.

Tell me when it's ready See what she does

In private testing. Not available for download yet.

Windows desktopAndroid phone appBring your own AINo account with us

The problem

Your AI forgets you. Your plans live somewhere else.

Most people who use AI juggle separate chat windows, apps and accounts. Each one forgets you between sessions, none of them knows your projects, and the answers stay on the screen: you still have to go and do the work.

Your plans live in one app, your notes in another, your documents in a third, and the AI sees none of them. On your phone, it starts over again.

How Arla solves it

Talk plainly. She keeps the thread and does the work.

Arla sits on top of your projects, goals, tasks, notes, documents and day, and connects to the AI models you choose.

Connect what you have

Use models on your own computer, free cloud services, or the paid AI accounts you already have. Mix them freely, with usage limits.

Say what you want

One continuous conversation on your computer and your phone. Interrupt any time: a correction folds into the work in progress.

She does it, with an Undo

Arla makes the change instead of describing it, asks first when it matters, and everything she changes can be taken back.

"Add a goal to read 12 books this year." "What should I focus on today?" "Switch to light mode." "Remind me to call the bank at 3."

What Arla does today

Everything here is built, tested, and running now.

No roadmap promises. These features work today on Windows and Android.

Conversation and models

One conversation on computer and phone, including what you say while the computer is off.
The best available model answers. If one fails, another takes over, and each reply shows which model answered.
Works with Ollama, LM Studio, OpenAI, Anthropic, Google Gemini, Groq, OpenRouter and any OpenAI-compatible service.
Voice: speak and hear replies. On the desktop, speech runs on your own computer.
Brain, Direct chat and Relay for power users: chain models to draft, critique and polish.

Plans, goals, tasks and the Timeline

Arla asks before she guesses: one short question for anything a new task or goal is missing.
Start does the right thing. Arla runs her own items; for yours, she shows just what you need to do.
Repeating items (daily, weekly, monthly) keep their history, and a missed turn says Missed.
A Timeline with Now, a clear Overdue mark, and bars that show how long each item runs.
Checks and routines: one-click project tests, and scheduled jobs on local models that cost nothing.

Today and your day

Today, Week, Month and Year calendars across all your projects.
Shrink to "Next up": the next three things, each with Start.
My day: set your hours. Work stays in work time, personal stays yours, and what didn't fit is listed.
She never moves an item you gave a time yourself.
Learns gently: she suggests a change only after a steady pattern, says why, and asks once.

Memory, Library and Research

Memory keeps a small, ranked picture of what matters, and tidies itself.
Library: your notes, PDFs and code, searchable by every model. Secret and backup files are left out on purpose.
Bring your history: import chats from ChatGPT, Claude and Gemini.
Research answers with numbered sources, labelled by publisher, saved for later.
Inbox: drop in an idea, link or file, and Arla files it.

Personas, Deck and more

Personas with their own model, voice and instructions: Code Buddy, Muse, Critic, Scout, or your own.
Deck: five chats side by side. Ask several models the same thing and compare.
Sticky notes on a board, on Home, or always on top. Also on the phone.
Image Studio: generate and save pictures with a connected service.
Nine themes, including a light one, and a Home built from the panels you choose.

The phone app

The same conversation, paired to your computer over your own network, on an encrypted link.
Works without the computer: an on-device AI model, fully offline once downloaded, or a cloud service you connect.
Messages sent offline sync when you're back.
Your own Home with Today, Timeline, goals, notes and personas, plus a quick orb.
Voice that can stay on the phone, and backups saved to your computer.

Everything Arla can do

The complete list.

Every feature below is in the app today, organised the way the app is.

Talking with Arla

One conversation across your computer and phone, including turns made on the phone while the computer is off. Long conversations stay quick: Arla keeps a short running summary instead of re-reading everything.
Your strongest connected model answers. If it fails, the next one takes over; with none reachable, a local model answers and any changes wait. Each reply shows the model that actually answered.
Talk over her: a correction mid-reply folds into the work in progress, and an unrelated request waits its turn.
Asks first before an action your words didn't request or work she proposed herself, and learns your phrasing over time.
Voice: speak and hear replies. On the computer, speech recognition and the voice run locally.
The orb: a movable orb opens Arla anywhere in the app; drag it out of the window for a floating orb with its own chat pop-up.
Bigger jobs: for a project-sized request, Arla drafts a plan with her team for you to review before anything runs.

Arla on your computer

Changes things for you: settings, pages, Home layout, projects, personas, your Inbox.
Power: sleep, lock, restart or shut down, always with a countdown you can cancel; wake the PC at a set time.
Reminders and requests scheduled for later ("at 9, summarise my inbox"), each one cancellable.
Hands for the AI: work with files in your workspace, run Python, read public web pages, and look at the screen when you ask.
Coding help: hand work to Claude Code on your own Claude Code sign-in, and open a PowerShell, Command Prompt or Git Bash terminal inside Arla.
Approvals: when an action needs your OK, approve it from any paired device; file edits and commands are approved on the computer only.

Many minds

Brain: a direct chat with your strongest connected model.
Deck: five chats side by side, to compare models or personas on the same question.
Direct chat: one conversation pinned to one model of your choice.
Relay: chain models step by step: draft with one, critique with a second, polish with a third.
Personas: Assistant, Code Buddy, Muse, Critic, Scout and Caretaker to start, each with its own model, temperature and instructions. Make your own with a full character sheet and icon.

Projects and action items

Projects each hold goals, tasks, plans, checks, routines, an Inbox, a Timeline and a Work log, and can point at a folder on your computer.
Clear stages: Not started, Needs your input, Running, In progress, Done. Start runs Arla's items and shows you just what to do on yours.
Details that matter: who it's for, priority, start and due dates, a time and length, and repeats (daily, weekly on chosen days, monthly or every few days, with an optional end).
Arla asks one short question when a new item is missing something it needs.
Manage in place: edit any card (even when finished), delete with Undo, and run a finished plan again while keeping its results.
A plan keeper re-paces your plans from real progress (finished work, saved files, passing checks) every couple of hours.
Undo for everything Arla changes; large batches wait on a review card.

Timeline, Today and your day

Timeline: a Now range around today, one Overdue mark, bars from start to the next checkpoint, several open cards, and drag to move a due date.
Today panel: Today, Week, Month and Year calendars across all projects; it shows more as it grows and shrinks to "Next up" with one click. Click it to open its own window.
My day: wake time, work hours and wind-down; each project marked Work, Personal or Both. Items are placed in fitting slots, and anything that doesn't fit is listed.
Gentle learning: only after a steady new pattern, she says why and asks once. She never moves a time you set.

Capture and notes

Inbox: drop in text, links and files (or a photo on the phone) and Arla files each one; a quick-capture window is always a click away.
Sticky notes: a Notes board, notes pinned to Home, and an always-on-top note; also on the phone, with offline edits that sync later.
Routines: small scheduled jobs on local models only, so they cost nothing: Morning focus, End of day summary and Weekly goal check to start.
Checks: one-click tests that confirm a project still works; a command is always shown to you first.
Work log: a live board of project work and of replies by service and model.

Memory, Library and Research

Memory: a small, ranked picture of you and your projects that tidies itself; edit, pin or delete anything.
Library: notes, PDFs, Word documents and code from folders you choose (and each project's folder), searchable by every model. Backup, signing and secret files are always skipped.
Bring your history: import chats from ChatGPT, Claude and Gemini; they become searchable notes, and the lasting facts go to memory.
Research: answers with numbered sources, labelled by publisher; check a source, save research with tags, and search it later. It uses your connected models' web search, with Tavily as a backup.

Workshop

Models: download and manage local models (kept on your graphics card so replies stay fast), connect cloud services, and set usage budgets.
Skills: small, pre-checked recipes for common jobs. Arla writes new ones as she learns, and each must pass its own tests before it's saved. Switch any off, export or import them.
Specialists: tiny local models that each do one job fast, taking routine work off the bigger ones.
Tools: connect other tools (MCP servers), start and stop them, and approve what each may touch.
Image Studio: generate pictures (Pollinations, Google Gemini, Hugging Face), search free images (Openverse, Pexels), and save them with their licence labels.
Test Arla: one button runs a short self-check with your own models and reports what worked.

The phone app

The same conversation and projects as your computer, paired over your own network on an encrypted link.
Works on its own: download an on-device model and chat fully offline, or use a cloud service whose key is on the phone (OpenRouter sign-in included). Offline messages and edits sync later.
Your Home: a Daily Brief, quick capture, chat, the Today calendar with My day, goals, Timeline, personas, notes, Inbox and health panels; it can mirror your computer's Home and accent colour.
Arla runs the app from chat: "open my notes", "switch to my work project", "add a goal".
A team on the phone: watch specialists work on a bigger request, step by step.
Camera to Inbox, voice that can stay on the phone, a quick orb, and backups saved to your computer.
What it's worth: an estimate of the time and money Arla saved you, with its assumptions shown.

Make it yours, and stay in control

Nine themes including a light one, accent colours, rename the app's labels, and a Home built from the panels you choose.
Control page: what Arla may do, what she did, which model answered, what she's learned, and what that learning costs in memory and disk.
Devices: pair phones, see what's connected, and refuse older, less secure links.
Backups: manual, nightly and before every update, with restore; phone backups kept separately.
Pop Home panels out into their own windows, and they reopen where you left them; full keyboard access; spelling suggestions on right-click.
Stays out of the way: closing the window keeps Arla in the system tray.

Works with

Ollama (local)LM Studio (local)On-device models (phone)OpenAIAnthropicGoogle GeminiGroqOpenRouterAny OpenAI-compatible serviceTavily (research)Claude CodeMCP tools

Privacy and control

Yours, on your devices.

There's no account with us, and none of our servers ever sees your conversations. A cloud model only gets a request when you've connected that service yourself.

Stored on your devices. No advertising, analytics or tracking.
Asks first before an action you didn't request, and before any command runs on your computer.
Undo for everything Arla changes. A large batch waits for your OK.
See what happened: which model answered, what Arla did, and what she's learning.
Backups: manual, nightly and before every update on the desktop; phone backups to your computer.

Read the privacy policy

What makes Arla different

Not another chat window.

One thread

One conversation follows you between computer and phone.

It acts

Arla changes things for you, and you can take it back.

Your models

Local, free and paid services side by side, with limits.

Your time

A day that keeps work in work hours, and learning that asks first.

Your data

It lives on your devices, not on our servers.

Platforms

Windows and Android, working together.

Windows desktop

Windows 10 or 11. Optional: Ollama (free) for AI models on your own computer. Without it, Arla uses the cloud services you connect.

Android phone app

Pairs with your Windows computer for the same conversation, projects and day, and keeps working on its own when the computer is out of reach.

Where things stand

Not available yet: Arla is in private testing. Leave your email and we'll tell you when it's ready.
Pricing hasn't been decided.
Windows and Android only for now: no Mac, Linux or iPhone.
The Windows installer isn't code-signed yet, so Windows may show a SmartScreen warning. The Android app isn't in the Play Store yet.
Arla's actions use a connected cloud model. Without one she still answers with a local model, and changes wait until a cloud model is back.

Questions

Good to know

Do I need to pay for AI services?

No. Arla can use free cloud services and models running on your own computer. If you already pay for an AI service, you can connect it too, and set usage limits so nothing runs away.

Does Arla send my data to you?

No. There's no account with us, and we have no server that receives your conversations. Your records stay on your computer and phone. A cloud AI service only receives a request when you've connected it. See the privacy policy for the details.

Can Arla change things without asking?

She asks first before an action your words didn't request, and before running a command on your computer. Changes she makes in one go are saved as one step you can undo, and a large batch waits for your OK.

Does it work offline?

The phone app can answer with an on-device model when it's offline, and your messages sync when you're back. On the computer, Arla can answer with a local model; actions wait until a connected cloud model is available.

When can I get it, and what will it cost?

Arla is in private testing, and pricing hasn't been decided. Email us and we'll let you know when it's ready.

Less switching. Less forgetting. More done.

Be the first to know when Arla is ready.

Tell me when it's ready