Lumen Agent · your control centre

The front door to every Lumen tool.

Give it a goal — “pull last quarter's numbers into a sheet, chart them, and draft a one-page summary” — and it runs the whole job, step by step, in the real editors, and hands back files you can keep changing.

Included on every plan, including free. Agent runs draw on your monthly AI credit allowance; your plan sets the engine and the allowance.

Ten tools, one control centre — hover to hold the orbit, click any tool.

Try the routing

One message in. Watch it pick the tools.

You never choose an app. Pick any goal below and watch which editors and services Agent routes it through — and what you're left holding at the end.

Anatomy of a run

You describe the outcome. It does the steps.

Nothing happens invisibly. Every run is a plan you can read, a timeline you can watch, and a set of files you can open — with your approval standing between the plan and anything consequential.

  1. 1 · Say what “done” looks like Plain language, not commands. Type it, dictate it, or start it from your phone. Attach files, paste an image, or point it at a folder — it will ask for what it still needs.
  2. 2 · It writes a plan Before touching anything it decides which sources to read, which editors to open and in what order. You see the plan first, and you can change it or stop there.
  3. 3 · You approve what matters Reading your own library is routine. Renaming, organizing, overwriting, sending an email, publishing a link or spending beyond a budget is not — those stop and wait for you.
  4. 4 · It works, in the open A live timeline shows each step as it happens: the tool it opened, the file it wrote, the search it ran. Cancel mid-run or retry a single failed step without starting over.
  5. 5 · You get real files Not a transcript to copy from — an actual sheet, deck, document, board, form or app in your library, editable, with a checkpoint saved before every AI revision.
What it can reach

It works with what you actually have.

Six sources of context, each turned on by you. Nothing here is enabled quietly, and every one of them can be switched back off.

Your library

Searches every PDF, sheet, document, board, deck, note, form and app you already have, opens the right one, and cross-references between them.

Files on your computer

Reads and writes inside the folders you grant it — including watching a folder so new arrivals get handled the same way every time.

The web

Searches the live web and fetches specific pages when a job needs current information, and tells you which sources it used.

Your screen

Looks at what you're looking at when you ask it to, so “fix the alignment on this slide” means the slide in front of you.

Connected services

Google Drive, Gmail, Calendar and YouTube — each approved separately, with read and write kept apart, and revocable at any time.

Your own tools, over MCP

Point it at a Model Context Protocol server and its tools join the same run, used alongside everything above.

What it can make

Every tool in the hub, on your behalf.

Agent is not a separate app with its own file format. It drives the same tools you use by hand, so anything it makes is something you can keep editing.

Autonomy with a handbrake

Let it work. Keep the last word.

An assistant that can open your files, spend credits and send email needs limits you set rather than limits it chooses. These are those limits.

Permission presetsDecide up front how much it may do without asking — from “check with me on everything” to “run the whole thing”. Change it mid-run.
Sensitive-action approvalsRenaming, reorganizing, overwriting, deleting, emailing and publishing always stop for a yes, whatever the preset.
Credit & step budgetsCap what a single run may spend and how many steps it may take. It stops at the ceiling and reports rather than quietly continuing.
A live, readable timelineEvery tool call, file touched and search made, in order, as it happens — not a summary written afterwards.
Cancel and retryStop a run at any point. Retry one failed step instead of re-running the job and paying for it twice.
A checkpoint before every AI editLumen saves the state a document started from, so an unwanted revision is one click from undone.
Local files stay localOrdinary editing happens on your disk. Content leaves the device only for the cloud step you asked for, and only what that step needs.
RecoveryAn interrupted run comes back where it was — closing the laptop mid-job doesn't lose the work already done.
Ways to reach it

Type it, say it, or start it from your pocket.

Voice, both ways

Dictate a goal instead of typing it, and have the answer read back. Useful when your hands are on the document rather than the keyboard.

Start it on your phone

Lumen Mobile runs a quick cloud Agent chat on its own, or hands the turn to your paired desktop — which has your files — and streams the result back into the same conversation.

It remembers

Preferences and past work carry between conversations, so the way you like proposals structured or sheets formatted doesn't need re-explaining every time.

Which engine runs it

Depth is a setting, not a guess.

Your plan decides how deeply Agent reasons. Short, mechanical jobs finish on the fast engine; long autonomous runs across many files are what the deeper engines are for.

Lumen FastQuick, high-volume steps. Free and up.
Lumen ProfessionalReal document work and long context. Plus and up.
Lumen ExpertReasons, then checks itself. Pro on demand · always-on from Max.
Lumen UltraMaximum depth, long autonomous runs. Ultra plan only.

See what each engine does in every tool →

Every documented capability

The whole list, in one place.

Multi-step planningLibrary searchAttachmentsImage understandingWeb searchFetch a URLScreen contextLocal file accessFolder watchingDocumentsSheets & dashboardsReportsSlidesDesign & videoFormsAppsAutomationsData transformationPython sandboxGoogle DriveGmailCalendarYouTubeMCP toolsConversation memoryVoice in & outMobile handoffLive timelineRecoveryPermission presetsSensitive-action approvalsCredit & step budgetsCancel & retryCross-document referencesCheckpoints before AI edits
FAQ

Lumen Agent questions, answered.

Will it change my files without asking?

No. Reading your own library is routine, but anything consequential — renaming, reorganizing, overwriting, deleting, sending an email, publishing a link — stops and waits for your approval, whatever permission preset you're using. Every AI revision also saves a checkpoint of the state it started from, so you can put a document back exactly as it was.

Do my files get uploaded?

Ordinary editing happens on your own disk. Content leaves your device only for the cloud step you asked for — an AI action, a share, a connector call — and only the content that step needs. Local files stay local unless you choose otherwise.

What does a run cost?

Agent runs draw on the same monthly AI credit allowance as the rest of Lumen. Every account gets free credits each month, and you can cap what a single run is allowed to spend. When a run reaches your budget it stops and reports instead of continuing.

Can it use my Google account?

Only if you connect it, one service at a time. Drive, Gmail, Calendar and YouTube are approved separately, read and write permissions stay apart, sends and publishing remain reviewable, and you can disconnect from Lumen or revoke from your Google Account at any time.

Can I use my own tools?

Yes — connect a Model Context Protocol (MCP) server and its tools become available in the same run, used alongside your library, the web and the Lumen editors.

What happens if a step fails?

The run stops at that step rather than guessing past it. You can retry that single step, change the plan, or cancel — the work already completed stays. An interrupted session comes back where it was.

Download

Give it the outcome. Keep the files.

Free to download, free to use. Every account starts with monthly AI credits.