A desktop AI agent that reads your files, runs your tools and writes your code — on your own API key, at your provider’s price.
Free desktop app · Your own API key · Cloud free during the beta
Everything in More AI happens in one of three modes. Switch whenever you like — and as you type, the app quietly suggests the one that fits.
Ask, think out loud, draft. A fast conversation with any model you’ve connected — nothing touches your files.
Point it at a folder and get real work done: it reads and edits your documents with visible diffs, and builds real Word, Excel and PowerPoint files.
A coding agent on your repository: it plans, edits across files, runs your tests and shows every change — with your approval at every step.
The same app, three different jobs. Pick the one that looks like yours.
Point the Code agent at your repository and it plans, edits across files, runs the test suite and shows every diff before anything lands. Because the app does the planning and the checking, an inexpensive model handles the routine work — a full day of agent time costs cents rather than a seat.
Give Cowork a folder of notes, contracts or research. It reads everything, folds the findings into your outline and writes real .docx, .xlsx and .pptx files you can send on. The folder never leaves your computer.
Schedule a morning brief, a weekly competitor scan or a long refactor. With the cloud switched on the run continues while your computer is off, and it is waiting as a conversation you can open from any device.
A warm, focused interface that gets real work done — chat, files, code, scheduled tasks and search — and stays out of your way during long sessions.
Double-tap Option (or press ⌘⇧Space) and a quick box appears on top of whatever app you’re in. Ask a question, attach a screenshot, get the answer right there — then carry on, or continue the thread in the full app.
Schedule the assistant to work on its own — a morning brief, a weekly check, a recurring report. Each run is saved as a conversation you can open whenever you like.
Long runs across Chat, Cowork and Code keep working in a background center while you do something else. Watch progress, open the transcript or stop a run — all from one place.
One shortcut searches across every chat, project and coding session at once — by title or by what’s inside — so nothing you’ve done ever gets lost.
A real browser lives inside the app. Preview your local dev server, browse the web, then screenshot and annotate a page — or point at a single element — and hand it straight to the assistant to look at.
Let the assistant look things up on the web and gather sources — pulling back image galleries, maps and more. Keyless search engines are built in, or connect your own premium one.
Built-in dictation turns speech into text in any field — no API key needed — and can tidy up the transcript with AI before it lands. Point it at a local model to keep it fully private.
The assistant remembers durable facts and preferences across conversations, kept as a plain file on your machine that you can read, edit or clear whenever you like.
As you type, the assistant quietly offers a one-click card when another mode or a command fits your request better — no menus to learn, and never more than one at a time.
It can read and write files, run commands, search the web, schedule routines and connect to your other apps — and you can turn any capability off in a single click.
Add a connector and its actions become available to the assistant in every mode — update a Notion page, read a Linear issue, look up a Stripe customer, open a pull request. Check the list for your own stack.
Adobe Workfront, Affinity, AgentMail, Ahrefs, AirOps, Airtable, alphaXiv, Amplitude, Apify, Apollo.io, Asana, Astro Docs, Atlassian, Attio, Base44, BioRender, Blockscout, Box, Brevo, Brex, Calendly, Campfire, Canva, Circleback, Clarify, Clay, Clerk, ClickUp, Close, Cloudflare Docs, Cloudflare Workers, Cloudinary, CoinDesk, Common Room, Consensus, Contentsquare, Context7, Coralogix, CourtListener, Craft, Crypto.com Market Data, Customer.io, Databricks, Datadog, DataHub, Day AI, dbt, DeepWiki, Descript, DevRev, Digits, Dovetail, Dropbox, Dynatrace, Egnyte, Elicit, Enterpret, Eraser, Exa, Excalidraw+, Fathom, Fellow, Felt, Fever, Figma, Fireflies, Gamma, GitHub, Glean, Globalping, GoCardless, Grafana Cloud, Grain, Granola, Guru, Gusto, Harmonic, Hex, Honeycomb, Hugging Face, HyperFrames, IFTTT, incident.io, Instacart, Intercom, Jam, Jotform, Kiwi.com, Klaviyo, Krisp, Linear, Lorikeet, Lucid, LunarCrush, Lusha, Magic Patterns, Mailchimp Transactional, MailerLite, Make, Mem, Mercury, Mermaid Chart, Metabase, Metaview, Microsoft Learn, Mintlify, Miro, Mixpanel, monday.com, Monte Carlo, MotherDuck, n8n, Neon, Netlify, Nimble, Notion, O’Reilly, Omni Analytics, Open Targets, OpenRouter, Otter.ai, Outreach, PagerDuty, PandaDoc, Parallel Search, PayPal, Peec AI, Pendo, Plaid, PlanetScale, PostHog, Postman, Prisma Postgres, Profound, Pylon, Qonto, QuickNode, Ramp, Razorpay, Read AI, Replit, Resend, Rillet, Salesloft, Sanity, Scite, Semgrep, Semrush, Sentry, Shopify Storefront, Sigma, Similarweb, Smartsheet, Snowflake, Socket, Sourcegraph, Spinach AI, Square, Stack Overflow, Strava, Stripe, Stytch, Sumble, Supabase, Superhuman Mail, Supermetrics, SurveyMonkey, SwaggerHub, Tableau, Tavily, ThoughtSpot, Todoist, TomTom Maps, Twilio Docs, Unthread, Unwrap, Vercel, Webflow, Whimsical, Windsor.ai, Wix, Wolfram, WordPress.com, Workable, Wrike, Zapier, ZoomInfo
Paste a URL and its tools show up alongside the built-in ones, in Chat, Cowork and Code alike.
Install from a .mcpb file — the same format Claude Desktop extensions use — and it runs on your machine, offline included.
Switch any connector on, off or down to read-only, and approve what the agent does with it.
The catalog is a shortcut, not a boundary: anything that speaks MCP can be added, and each connector keeps its own permissions.
More AI talks to your provider straight from your computer, with a key you supply. No middleman and no markup: you pay OpenAI, Anthropic, Google or any compatible provider at cost, take an inexpensive model for everyday work and a top-tier one when it matters — and you can see what any task cost. Would rather not manage a key? Buy credits from us: the same models, one bill, a small markup over provider cost.
OpenAI, Anthropic, Google, Mistral, Groq, DeepSeek, OpenRouter and more — or a model running fully locally via Ollama.
More AI is a native desktop app. Sign in and the same chats and projects follow you into your browser, the Chrome extension and your phone — one account, one workspace.
The full experience — Chat, Cowork and Code — running on your machine. Free, no sign-up.
The assistant on the page you’re already on: it reads what’s on screen, takes context straight from the tab and hands it to an agent.
Your synced chats and projects in any browser at app.more-ai.net. Sign in with just your email — no password.
The web app already works on your phone today, and native mobile apps are next.
Everything in More AI runs locally by default. Switch on the cloud and long jobs move to our machines: the agent keeps planning, editing and running code while your computer is off — and you follow along from any device.
Hand it to the cloud and shut the lid. Background and scheduled runs — a morning brief, a weekly report, a long refactor — carry on and are done and waiting when you come back.
Cloud runs happen in an isolated environment with its own filesystem and network, so an agent can install packages, execute commands and test its own work without touching your computer.
Chats and projects stay in step across your computers and open in the browser, the Chrome extension and on your phone. Start at your desk, check progress from the train.
Shared workspaces, shared connectors and one budget across the team with a limit per member — so you can see where the spend goes without taking control away from anyone.
Cloud runs, sync and every feature of the app are open to everyone at no cost for as long as the beta lasts — no card, no quota. Tokens still go through your own key. The subscription starts only when the beta ends, and we’ll say so well in advance.
The subscription pays for the cloud — the machines your agents run on and the sync that keeps the results in reach. Tokens are never bundled in: they stay on your own key, at your provider’s price, under a budget you set.
The account already works — sign in with a code sent to your email, no password.
The same capability, a very different deal on privacy, price and control.
| More AI | Cloud apps | |
|---|---|---|
| Where your work lives | On your machine; cloud is opt-in | On their servers |
| What you pay for | The app is free; tokens at provider cost | A seat, with usage baked in |
| Who sets the limits | You do — your key, your ceiling | Their plan does |
| Where it runs | macOS · Windows · Linux · web · phone | Varies |
| Works offline | Yes, with a local model | No |
| Choice of models | Any provider, or fully local | One vendor’s models |
Short answers to what usually decides it. The docs go deeper.
The desktop app is free — Chat, Cowork, Code, connectors and local models, with no sign-up. The paid layer is the cloud: the machines your agents run on when your computer is off, and the sync that keeps the results in reach. While the beta lasts, that is free for everyone too. Tokens are never bundled — they always run on your own API key at your provider’s price.
Not for the desktop app: install it, add a key and start working. An account is only needed for the cloud — syncing across your computers, opening your work in a browser or on your phone, and runs that continue without you.
Conversations, project files and context live in local storage on your computer. Nothing is synced anywhere until you switch the cloud on yourself, and then only the sessions you start there.
Any provider you hold a key for — OpenAI, Anthropic, Google, Mistral, Groq, DeepSeek, OpenRouter and other compatible APIs — or a model running fully locally through Ollama. You choose the model per conversation, and you can change it mid-thread.
You pay your provider directly, at their price, with no markup from us. The app does the planning, the tool calls, the file edits and the self-checks around the model, which is why an inexpensive model handles most everyday work and the expensive one is only needed where it counts. Every task shows what it spent.
Yes, with a local model through Ollama. The app, your files and your history need no connection — only calls to a hosted provider do.
Native desktop builds for macOS, Windows and Linux, a Chrome extension in beta, and the web app at app.more-ai.net in any browser, including a phone. Native mobile apps come next.
Every release is submitted to VirusTotal before it is published, and the result — which engines scanned it, the file hashes and a link to the report — is on the security page. You can verify the file rather than take our word for it.
Free to download, yours to run. Your keys, your computer, your budget.
Grab the build for your platform and open it. macOS, Windows and Linux are all first-class.
Paste in a provider key — OpenAI, Anthropic, Google and more. It’s stored locally on your machine, and used only for the provider you added it for.
Open a project and start a conversation. The assistant works with your files, and nothing leaves your computer until you switch on the cloud.