How can we help?

An operating system that turns AI agents into an office you can see and manage. Every agent has a seat, every output lands on one board, and nothing goes out without your approval.

You open it in a browser as the Office: twelve pages, each with one job.

On your own machine or in the cloud. On your machine, the engine runs locally with no third-party packages, and your workspace is a folder of plain files you can open, back up or move.

Start small: pick one desk and a few routines you already do every week. Install or create a seat, give it a task, and approve its first output from the Inbox. Add more once it proves itself.

Overview, Inbox, Chat, Tasks, Outputs, Agents, Workflows, Progress, Store, Database, Activity and Map. They are grouped as One board, Team and Workspace.

In the Inbox: reviews, decisions and problems in one queue. Overview also shows how many items are waiting, alongside AI health and today’s deliveries.

Every deliverable lands on the Outputs board, where you grade it. On disk they are files in artifacts/.

A seat is an AI agent with a job, a place on the org chart and a record of its work. Each one reports whether it is working, idle or blocked, and has a profile on the Agents page.

From Tasks: assign work once or on a schedule. Recurring work lives in Workflows, which show each routine’s steps, schedule and last run.

Yes, in Chat. A live conversation runs under the same rules as a scheduled run, so anything external still waits for your approval.

Install a seat template from the Store, or save one of your own as a template.

Every send, publish, CRM update or payment. Agents can prepare anything. Nothing goes out until you say yes.

Approve and the engine carries out the action. Reject and nothing goes out. Answer a decision and your choice can be saved as a standing preference the agent reads on its next run.

No. The Gmail connector only drafts to you. An agent never sends to others. Sending is an action the engine takes only after your approval. In Chrome, an agent can fill a form but stops before submit.

Gmail (drafts to you), Google Calendar (reads your schedule), Google Drive (files and documents), Notion (trackers, drafts and databases), Claude in Chrome (fills forms, stops before submit), HubSpot through a CRM queue, and web research.

HubSpot updates wait in a queue and go through the same approval as any other outward action.

Yes. The home page lists what is connected today and what is planned. Planned tools are only named. Nothing is claimed until it ships.

When you answer a decision, your answer can be saved as a preference. Preferences are kept in knowledge/preferences.md and every agent reads them before it works, so you correct an agent once, not every run.

In plain files in your workspace. Approvals are in state/approvals.json, every run is logged in state/runs.json, each seat reports to state/agents/<seat>.json, and outputs are in artifacts/.

Yes. The Database page shows your OS data and your own tables, boards and calendars, and Activity shows every run and delivery by day.

Three: a success rate of 95% or higher (the share of runs that finish), a first-try rate of 70% or higher (the share of outputs you approve first time), and a cost per finished item that stays low (at or under the middle cost of all finished work).

Over 25 reviews in 30 days, it has to meet the success and first-try bars and an 85% review score. Running more never counts. Only finished, reviewed work does.

Eight stages your company climbs, each earned from reviewed work. Progress shows the next stage and what unlocks it.

Four: organization, person, community and project. Every workspace runs the same engine.

Yes, on the Several workspaces plan: one company or several, switched from the sidebar, with templates shared through the Store.

Pricing is quoted, not published. It follows how many workspaces you run, and where. Request pricing and a person replies.

Open the Inbox: problems wait there alongside reviews and decisions, so start with anything from that seat. The Agents page shows each agent as working, idle or blocked.

Reject it, or answer with what you want instead. Save the answer as a preference and the next output comes back the way you asked.

Write to team@manelink.com and a person replies.