Opt-in. Off until you switch it on.

One main agent plans, hands work to specialists and comes back to you. You watch it happen, approve what matters, and every line they write is signed. Off until you switch it on.

Join the betaHow control works

based

Mission control

Inbox3

Agents4

Projects6

Schedules2

Graph

Monday, Mission control

Good evening, Lena

Based main agent, 01:47

The Acme proposal is drafted and your investor update has fresh numbers. One thing needs you: approve two edits in the proposal.

Working on

Acme proposal

18 %

Main + 2 agents, Deep

Thesis ch. 2 sources

52 %

Researcher, Claude

Investor update numbers

30 %

Coder, local model

Budget 2026 review

70 %

Waiting for your approval

Live activity

22:41Main split “Acme proposal” into 4 tasks

22:39Writer turned the meeting into action items, 5 owners

22:37Researcher compared 4 vendors, table ready

22:35Main wrote the project update, 3 changes since you looked

22:33Coder updated the formula in Budget 2026

22:31Night shift filed 7 voice notes into Inbox

Prepare the board meeting and keep the project page updatedMainAutoAsk before editsPrivacy check on

Live demo with sample data. Yes, it keeps going.

Proactive

It comes to you.

You don't check on agents. They check in: when something is done, when they need a decision, and every morning with a briefing. On your Mac, on your phone, in your inbox.

Morning briefing 07:302 deadlines today, 1 approval waiting, 3 pages changed overnight.
Done 11:02Competitor table added to Projects/Launch. 14 sources linked.
Needs you 16:58Approve the Acme proposal draft? Two paragraphs changed.
Night shift 23:10Filed 7 voice notes and updated 3 client pages.

Native to your databases

Your morning briefing is a database row.

Agents don't live in a chat window next to your work. They read and write your databases like you do. Every briefing, run and decision becomes a row you can filter, relate and search, so the history of your work builds itself.

Morning briefings

DateSummaryDeadlinesNeeds youLinked
Tue, Nov 10Acme proposal due Friday. Thesis sources complete. Budget review moved.21 approvalAcme, Thesis, Budget
Mon, Nov 9Investor update drafted. Northwind kickoff notes filed.1noneInvestors, Northwind
Fri, Nov 6Offsite booked. 3 new leads from the website form.02 decisionsOffsite, Leads
Thu, Nov 5Lecture 4 summarised. Hiring post waiting for budget.3noneUniversity, Hiring
Wed, Nov 4Lumen API webhook live. Runs now logged.11 approvalLumen Labs, Logs

One session, everywhere

Terminal, cloud, phone. Same conversation.

Start a task from your terminal, let it keep running in the cloud sandbox when you close the lid, answer its question from your phone. Any AI can join through MCP. Every message and every result is saved to your workspace as it happens.

$ based ask "update the Acme deal"

Terminal

Talk to your workspace from the CLI. Works offline against your local copy.

→ continues in sandbox

Cloud sandbox

Close the lid. Long tasks keep running in an isolated sandbox, opt-in.

“Yes, send it Friday”

Phone

Get the ping, answer in one line. Same session, nothing to re-explain.

mcp://based

Any AI via MCP

Claude, ChatGPT, Cursor and others read and write your workspace, with your rules.

Cloud agents are opt-in and only read the pages you hand them, while they work. Based Cloud runs on EU servers and is end-to-end encrypted by release. The beta starts on the Mac with local storage and iCloud.

Control

You set the leash.

Four positions, per task. Change it any time, even mid-run.

Default

Work freely, check in before changing your pages.

Agents research, draft and organise in their own space. Before they touch your pages, they ask, once per change set.

Writer wants to replace 2 paragraphs in “Thesis ch. 2”.Show diffApprove

Orchestration

One main agent.
As many hands as the job needs.

You talk to one agent. It picks the models, spawns specialists, checks their work and writes the update into your project page. Like a chief of staff who never sleeps and always cites.

You Main agent Researcher Writer Coder Reviewer Projectpage, briefinbox

Progress

Progress you can see, not promises you have to trust.

Projects, runs and tasks on one board. Cards move as agents finish. Click any card to see what was done, with which model, at what cost.

Queued1

Board meeting prep

Main, Deep

Running2

Call summary Acme

Writer, Fast

Citations check

Reviewer, Auto

Needs you2

Sources ch. 2

Researcher, Claude

Voice notes

Night shift

Done2

Budget formulas

Coder, GPT

Internship table

Researcher, Deep

Night shift

Go to bed. Wake up to progress.

Schedule agents to run while you sleep: file the day's voice notes, refresh the project pages, prepare tomorrow's brief. In the morning, the marker shows exactly what changed.

True story: Based itself is built this way. A night shift of agents picks tasks from the backlog, runs the tests and leaves a log by morning.

night-shift.log

Models

Any model. Your keys. No credits.

Connect the ChatGPT or Claude account you already pay for, plug in an API key, or drive the coding CLIs you already use. Pick per task: Auto, Fast, Deep or a specific model. Or run one locally.

ChatGPTClaudeClaude CodeCodexGemini CLIOpenCodeOpenAI APIAnthropic APIMCP for CursorMCP for ClaudeMCP for ChatGPTYour own serverChatGPTClaudeClaude CodeCodexGemini CLIOpenCodeOpenAI APIAnthropic APIMCP for CursorMCP for ClaudeMCP for ChatGPTYour own server

Skills

Teach an agent a routine once, like how you write client updates, and every agent can use it.

Memories

What agents learn about your projects, kept as pages you can read, edit and delete.

Daily brief

Deadlines, approvals and what changed overnight, saved as a row in your briefings database.

Answers with sources

Ask your whole workspace. Every answer links to the pages it came from.

Privacy check

Opt-in and planned: before a prompt leaves your device, a local check flags names, numbers and other private details.

Local models

Run a model on your own machine for anything sensitive. Nothing leaves the device.

They do the work.
You stay in charge.

Agents are optional, marked and on your terms. Join the beta for the Mac.

Join the beta