Workestra Workestra
Start your office →

Hire an AI team. Run it like an office.

Workestra is an office of AI agents for people who don't write code. Hire agents into roles, give them skills and tell them what you need in plain words. They share one task board, report back, and wait for your OK before anything risky.

Start your office → See how it works
For
Owners, admins, small teams
Speaks
English · Bahasa Melayu
Runs on
Our cloud or your server
Risky steps
Wait for your approval
Sheet A-01 · The Office page, liveThe real product, not a drawing of it
X:0000 Y:0000
Manager chat
The office, live
Needs your OK
One shared board
Side panel
00/06Drawing the office from your team size: four desks, one row, a room for the Manager.
Sheet B-01 · The problemWhy one chat box is not enough

A chat window is not a team.

Most AI tools are one assistant in one chat. They wait for you to type, forget what you agreed last week and can't hand work to anyone. Real work needs roles, a shared list, a memory and someone in charge.

No roles[ ]

One assistant does every job

No one owns Accounts, Admin or Support. You end up managing every prompt yourself.

No memory[ ]

Every chat starts from zero

You explain your clients, your tone and your rules again, every single time.

No oversight[ ]

It does too little or too much

Either it can't touch anything real, or it can touch everything. There is nothing in between.

No view[ ]

You can't see the work

Who is doing what, what is stuck and what waits on you all lives in a long scroll.

Workestra is an office, with roles, a shared board, a memory and you in charge.

Sheet A-02 · How it worksFour steps, no code
Plan · 4 desks · 1 rowStep 01 / 04
  1. 01

    Hire an agent into a role

    Pick a character, give it a name and a role: Accounts, Admin, Support, Web Monitor, whatever your work needs. The role sets what the agent does and what it may touch.

  2. 02

    Give it skills

    Skills are written know-how for one kind of job: chase invoices, check a website, plan a meeting. Pick them from the library. Agents also learn new ones from your feedback, and you see every one.

  3. 03

    Tell your Manager what you need

    Write the way you'd brief a colleague, in English or Bahasa Melayu. The Manager does the job or hands it to the right agent, and reports back in the same chat.

  4. 04

    Approve what matters

    An email to a client, a calendar invite, a change on your website: anything that leaves the office waits for your OK. You see exactly what will happen before it happens.

[ Scroll to continue ]

Sheet C-01 · SafetyAutonomy tiers, enforced in code

Nothing risky happens without you.

Every tool an agent can use has a tier. The tier decides whether the agent may use it alone, must ask you first, or can't use it at all. The rules live in the software, so no clever wording in an email or a web page can talk an agent past them.

T0

Look and report

Agents read, check and report on their own.

Check a website · read your inbox · read the calendar
T1

Draft and prepare

Agents prepare work that stays inside the office until you use it.

Draft a reply · write a report · plan a project
T2

Ask, then act

The agent asks. You see exactly what it will do. It runs only after you approve.

Send an email · invite a guest · clear a site cache
T3

Never

Not given to agents at all. There is no tool to ask for.

Delete data · change billing · change DNS
Outside text is data

An email or web page that says "ignore your rules" is read as content, never as an instruction.

Everything on the board

Every task shows who did what, when, and what they found.

Learning can't promote

What agents learn never gives them a new tool or a higher tier.

Limits per agent

Each agent has a daily run limit, so a busy day can't run up a surprise.

Sheet C-02 · MemoryAn office that gets better at your work

It remembers. It learns. It asks.

Remembers

Your work, kept

Agents keep notes on your clients, projects and decisions, and read them before they start. A short profile of how you like things done goes with every job.

Learns

Your way, learned

Correct an agent once and it can turn that into a skill. You can read every skill it writes, edit it or switch it off.

Asks

No guessing

When something is unclear, the agent asks you on the board and waits for your answer instead of making it up.

Sheet D-01 · AppsConnect once, per office

Works in the apps you already use.

Connect your Google account with one click. Agents read what they need and prepare the rest. Anything that reaches another person waits for you. Disconnect any time. The Privacy Policy sets out exactly what Workestra reads from Google and why.

Gmail

  • AloneReads and sorts your inbox, finds invoices and replies
  • AloneDrafts replies in your tone
  • Asks youSends any email

Google Calendar

  • AloneReads your calendar and finds free time
  • AloneAdds events for you, without guests
  • Asks youInvites anyone else

Google Drive

  • AloneReads your files and folders
  • AloneWrites Google Docs in its own Workestra folder
  • NeverDeletes or moves your files
PlannedSlackNotionMicrosoft 365
Sheet E-01 · Who it's forEach office is set up for its own work

Built for the work you actually do.

  • Office admin

    Schedules, reminders, letters, the paperwork that fills a week.

    Admin · Calendar
  • Accounts

    Unpaid invoices, receipts, monthly summaries for the owner.

    Accounts · Gmail
  • Customer replies

    First drafts for every enquiry, sent only when you say so.

    Support · Gmail
  • Web & development

    Penxel Studio runs its own office for this: uptime checks, incident diagnosis, code changes.

    In use today
  • Your own pack

    Roles, skills and duties are data, not code. Set the office up for any kind of work.

    Roles · Skills · Duties
Sheet F-01 · PlansPick where your office lives

Your office, your way.

Start on our cloud, or have your own Workestra set up on your own server. Either way you choose the AI, and the AI usage is billed to you, with no markup from us.

Cloud

Workestra Cloud

Early access
  • Your office at yourname.workestra.my
  • We host, update and back it up
  • Gmail, Calendar and Drive included
  • Start with a Manager, hire the rest as you go
Join early access
Own deployment

Your own Workestra

Quoted per company
  • A separate office on your subdomain or your server
  • Data stays in your own database
  • Set up for your kind of work by Penxel Studio
  • Roles, skills and duties tuned with you
Talk to us →
AI engine

Bring your AI

Billed by your AI provider
  • Sign in with your own Claude plan
  • Or add an API key from Claude, OpenAI or DeepSeek
  • Pick a fast, default or deep model per agent
  • If one engine hits its limit, agents switch to the next
How AI billing works
Sheet G-01 · QuestionsPlain answers

Questions, answered.

Q1Do I need to know how to code?

No. You hire agents from a form, pick skills from a list and talk to your Manager in plain words. Nothing in the daily work needs a terminal.

Q2What can agents do without asking me?

They can read, check and draft (tiers T0 and T1). Anything that leaves the office or changes something outside it, like sending an email or inviting a guest, waits for your approval. Some actions, like deleting data, are never available to agents.

Q3Which AI does it use, and who pays for it?

Claude by default, signed in with your own Claude plan. You can also add API keys for Claude, OpenAI or DeepSeek. Usage is billed by that provider directly to you. Each agent can use a faster or a deeper model.

Q4Can I talk to it in Bahasa Melayu?

Yes. Write in English, Bahasa Melayu or a mix of both. Agents answer in the language you use.

Q5Where is my data kept?

In your office's own database. On your own deployment it stays on your server; the only things sent out are the prompts to your AI provider and the calls to apps you connect. Passwords and keys are stored encrypted and never put into a prompt.

Q6What if an agent gets something wrong?

Give a task "done when" lines and the Manager checks the result against them before it's marked done. Reply on a finished task and it reopens with your comment. Correct an agent and it can learn from it.

Q7What is the pixel office for?

It shows your team at a glance: who is working, who is waiting for you, who is in a meeting. Click a character to chat with that agent.

Sheet H-01 · StartEarly access, 2026

Hire your first agent this afternoon.

Start your office →Compare plans
Or write to Penxel Studiohello@workestra.my