01 · Hello
Outpost
How it worksSecurityQuestionsSign in

Your personal agent · in Telegram and WhatsApp

Work, handled.One message away.

Outpost is an agent with a computer of its own. It researches, writes and runs code, handles your inbox and calendar, and keeps going on a schedule. You talk to it in Telegram or WhatsApp. It asks before doing anything other people would see.

Sign inSee how it works ↓
  • Nothing to install
  • Asks before it acts
SolvePactonline

Every weekday at 8, send me a brief of my day.21:14

Your agent wants toSchedule “Morning brief”Weekdays at 08:0021:14
AllowAllowedDeny

Done ✅ The first one arrives tomorrow at 08:00.21:15

Today

⏰ Morning briefThree meetings, the first at 10:00. Two emails need a reply today. Your pull request was approved overnight, ready to merge.08:00

Message

How it works

One message.Real work.

  1. 01

    You ask, in plain words.

    Message your agent in Telegram or WhatsApp the way you would message a colleague. No commands to learn.

  2. 02

    It does the work.

    On a private computer of its own it reads, searches, writes and runs code. Every step is logged where you can see it.

  3. 03

    You approve what leaves.

    Sending, posting, pushing: anything other people would see waits for your Allow or Deny. Then the answer arrives.

SolvePactonline

The login test is failing on main. Fix it and open a pull request.14:02

Working…

Read auth/session.ts

Ran npm test · 1 failing

Wrote auth/session.ts

Ran npm test · all passing

Your agent wants toPush to a remote repositorygit push origin fix/login-test14:04
AllowAllowedDeny

Fixed 🎉 The session expired a day early. The pull request is open and the tests pass.14:05

Message

AskDone

  • Every steplogged
  • Nothing sentunasked
  • Your keysunseen

Security

Keys the agentcan’t read.

Most agents are handed your passwords and trusted to behave. Outpost is built the other way round: the agent says where a secret belongs, and the server puts it there. The AI never holds it.

  • Used only on the site it was saved for
  • Each use needs your Allow
  • Removed from anything the agent gets back

$ vault_request “Bank”

agent wrote
password={{password}}
site check
bank.example · saved for this site
you
Allowed
server sent
password=•••••••••••• · filled in
model saw
{{password}} · HTTP 200

Signed in. The password never left the server.

A sealed room per task

Every task runs in a fresh, locked-down container holding your files and nobody else’s.

Used, never read back

What you save is encrypted. Not even you can display it again, so nothing else can either.

Yours to take or erase

Download everything stored about you, or delete it all, from Settings.

3,000+ apps

The tools you use,one tap each.

A directory of more than 3,000 apps, nearly 2,000 of them with ready-made actions your agent can use today. Connect one by signing in to it.

Then, for every tool, you decide

  • Run freelyfor tools that only read
  • Ask firstfor anything that sends or changes
  • Offhidden from your agent

Also in the box

Everything elseyou’d reach for.

  1. 01

    Schedules and reminders

    “Every weekday at 8, brief me on my day.” Say it once. It runs on its own and messages you the result.

  2. 02

    A memory you can read

    It keeps notes on how you like things done, and uses them in every conversation. Read them, fix them or clear them.

  3. 03

    A vault it can use but not open

    Save a login once. The agent can sign in with it, only on the site it was saved for, and never sees the password.

  4. 04

    Your tools, connected

    Gmail, Calendar, GitHub, Notion, Slack and a directory of 3,000+ apps. One tap each.

  5. 05

    A record of everything

    Each task keeps its steps, the files it changed and what it cost. Long answers in chat link to the full version.

  6. 06

    The door is always open

    Download everything stored about you, or delete your account and all of it, from Settings.

06 / 06

Questions

Asked before you ask.

Do I need to install anything?

No. Your agent is a contact in Telegram or WhatsApp, apps you likely have already, and this website is where you see its work and manage its tools. Nothing to download.

What can I ask it to do?

Research a topic and write it up, fix a bug and open a pull request, summarise documents you send, check your inbox or calendar once those are connected, or do any of that on a schedule.

Can it do things without asking me?

It works on its own inside its private workspace. Anything that reaches other people, such as sending an email, posting, pushing code or using a saved login, stops and asks you first. If you do not answer, it does not happen.

Does the AI see my passwords?

No. Passwords and keys are encrypted and stay on the server. When the agent needs one, it says where the password should go and the server fills it in. The password is never shown to the AI and never appears in a task's log.

Which apps does it work with?

The directory on the Settings page lists more than 3,000 apps, and nearly 2,000 of them already come with ready-made actions your agent can use. Any service that offers an MCP server can be added by its address too.

How do I sign in?

There is no password. Press “Sign in” and continue with Telegram or WhatsApp. In the chat that opens, press Start or send the message, then tap Sign in when your agent asks. The page signs you in by itself.

Your turn

Say helloto your agent.

Sign in

One message is all it takes to start.

OutpostPrivacyTermsSign in