You Your brief Human in the loop

◆

Orchestrator

Plans · Delegates · Reviews

Nina

#ui-ux

Kai

#dev-build

Elena

#research

Mia

#marketing

Jonas

#community

Sofia

#partners

Clara

#finance-legal

Noah

#trust-talent

A team of agents.

You send a task. It gets split up, handed to whichever agents are right for it, and comes back done. This is work you would otherwise hire a developer, a designer or a researcher to do. And it runs where you already work: Telegram and web chat today, Discord, Slack, WhatsApp or your own app on request. Borrow ours, or we teach you to build your own.

Live activity

/ 01 · How it works

You ask.
It gets done.

Most AI setups leave you doing the coordinating. This one does that part too. You still sign off on anything that leaves the building.

Step / 01

Say what you need

Normal words. Chat, voice note, ticket, whatever suits you. Nobody here needs to write prompts for a living.

Step / 02

It gets broken up

The orchestrator works out what the job actually involves, who should do which bit, and what has to happen before what. It is the only one holding the full picture.

Step / 03

Everyone works at once

Elena digs up the research while Kai is already building and Mia drafts the copy. That is where the speed comes from, not from any single agent being clever.

Step / 04

It comes back together

The pieces get checked against what you originally asked for. If two of them contradict each other, that goes back round before you ever see it.

Step / 05

You have the last word

Nothing goes out without someone looking at it. Send it back with a note and the whole thread stays intact, so nobody starts from zero.

/ What this means in practice

01 /

It doesn’t clock off

Send something at 11pm and it is usually waiting for you by breakfast.

02 /

You can see everything

Every step lands in a channel you can scroll back through. No mystery outputs.

03 /

Busy week? Fine

Ten tasks instead of two just means ten running at the same time. No onboarding, no ramp-up.

04 /

Nothing happens behind your back

Approvals, budgets and who can touch what: all yours to set. The model does not get to decide.

/ Real recording · nothing staged

Two minutes,
one job after another.

Recorded in the real team chat. None of it is staged.

Kai reads the open pull request, sums up what it changes, and says where the risk sits.

Clara turns one sentence into an invoice PDF. It only gets an invoice number after approval.

Clara drafts a twelve month agreement with real clauses, signature lines and the logo.

Elena researches the market, opens the sources, and says what she could not verify.

Mia writes the post, Nina supplies the image for it.

The draft gets rejected and rewritten. Nothing goes out without a yes.

This is Telegram, and that is only where we happened to record it. The same team runs in WhatsApp, Discord, Slack, a chat on your own website, or inside your own app. The chat is the door, not the product.

/ Watch it work

You say what you need

You"Build me a landing page for the product launch."

Elenaon it

Ninaon it

Kaion it

Miaon it

shares research on competitor pages

sends first layout to Kai

builds the page from Nina's layout

drops launch copy in for Kai

yourcompany.com/launch

ProductTeamContact

The launch you've been waiting for. Available now, try it risk free. Get started

✓Draft 1 is ready for your review

/ Another run · an invoice

You say it in passing

You"Clara, invoice Asta in the UK: 1x iPhone 15, 1,000 EUR, no VAT, as a PDF."

Claraon it

asks once for what is still missing

computes net, tax and total in code, not in its head

notes that a UK customer needs a line explaining the zero VAT

builds the PDF, in English

Invoice Draft, no number yet

Bill toAsta · United Kingdom

1 × iPhone 151,000.00 EUR

plus 0 % VAT0.00 EUR

Total1,000.00 EUR

DRAFT · invoice number only after you approve

✓Ready as a PDF. Say yes and it gets its number.

An agent writes to a customer

You"Quick email to our customer Daniel: tomorrow's meeting has to be 3pm, not later."

Jonason it

Todaniel@customer.com

SubjectMeeting time change tomorrow

Hi Daniel, small change: our meeting tomorrow is now at 3pm, not later. Please update your calendar. Thanks!

✓Sent to Daniel, no approval needed

/ Try it yourself

Click a sentence and see how an agent would respond.

Kai

/ 02 · The agents

Nine of them.
Each does one thing.

We tried the one big agent that does everything. It was worse at all of it. So each one here has a single job, its own channel and only the access it actually needs. A role here is not a label: if an agent is listed for a job, it has the access to carry that job out, not just to talk about it.

Leo, Kai, Clara, Elena, Mia, Nina, Jonas, Sofia and Noah

01 /

Orchestrator #core

Reads what you asked for, works out the steps, hands them out and checks what comes back. The only one that sees the whole job.

Planning Routing QA

02 /

Nina #ui-ux

Wireframes, interface concepts, design reviews. Spots the holes in a flow before anyone starts building.

Figma Flows Design systems

03 /

Kai #dev-build

Writes code, tests it, deploys it, clears out bugs. Works straight in the repo.

Code Tests Deploys

04 /

Elena #research

Market, competitors, users. Delivers summaries with sources, never bare claims.

Sources Synthesis Briefs

05 /

Mia #marketing

Positioning and copy. Picks up tone and audience from the brand context.

Copy Campaigns SEO

06 /

Jonas #community

Answers questions and keeps track of what needs to happen when. Whatever keeps coming up gets routed back to the product.

Support Moderation Signals

07 /

Sofia #partners

Works the inbox, reads the whole thread and drafts the reply it actually calls for.

Leads Outreach CRM

08 /

Clara #finance-legal

Invoices, receipts, quotes and contracts. Give her the facts in a chat and the finished document comes back, as a PDF or an editable file that opens in Word, in German or English. An invoice number is only assigned after you approve it.

Invoices Receipts Contracts PDF & Word Review

09 /

Noah #trust-talent

Checks out vendors and applicants, and keeps an eye on who has access to what.

Vetting Access Policy

/ 03 · Work with us

Coming soon, we're opening this up shortly.

Rent it. Learn it.
Or own it.

Use the agents as they are, or we build them onto your own systems.

All three start with the same call. We look at how you actually work and tell you which one fits. Sometimes the honest answer is none of them yet, and we will say that too.

Option 01 /

Just use
ours

You get a channel, you send tasks into it, work comes back. Nothing to install, nothing to configure on your end.

Live in days Monthly No setup

Option 02 /

Learn how
we did it

We open the thing up and show you how it is wired. Then we build your first agent together, on something you actually do every week. No slides.

1:1 or team On your process Recorded

Option 03 /

We build it
into your shop

Runs on your servers, on your data. We shape the agents around how your team already works, set it up, write it down and train your people. Then it is yours.

Full handover Your infra Support

/ Common questions

Q / 01

Does this replace my team?

No. It takes the repetitive stuff off their plate. The judgement calls still land with a person, and that person is on your side of the table.

Q / 02

What happens to my data?

With option three, nothing leaves your servers. With the other two we write down beforehand exactly what each agent can reach, and it gets nothing beyond that.

Q / 03

Do I need technical knowledge?

Not for option one. For option two you mainly need to know your own business well. Coding is not part of it.

/ 04 · Capabilities & stack

What they
can do.

Everything under “live” runs today, not on a roadmap. Below that is what you connect with your own keys, and what the whole thing runs on: boring, well understood pieces you can read, check and take over whenever you want.

/ What the agents do · live

Open pull requests Merge a pull request Run checks before merging Read and search any repo Ship a ZIP into a repo Inspect deploys and their state Read build logs List env variable names Request a redeploy Look up a Solana transaction Read program logs on a revert Wallet and token balances Identify an on-chain address Search the inbox Read a full email Reply grounded in the thread Send email Draft and post to X Research with sources Review a design or screenshot Generate images Understand voice messages Remind you before it matters Watch a deploy and report back Keep an address book Message a contact for you Daily health check Answer in your language Durable memory across chats Read-only agents for finance and trust

/ What you connect · your keys

Helius Solana RPC Webhooks Any REST API MCP Your own database Your own model key

/ Models & execution

Claude API Tool use / function calling Structured outputs Retrieval Sandboxed execution

/ Where they run · live

Telegram Web chat, no Telegram needed

/ Where they can run · an integration, not a rewrite

Discord Slack WhatsApp Business Your own app Any chat with an API

/ Documents · finished, out of a chat

Invoices Receipts Quotes Contracts and agreements As PDF As an editable Word file German or English

/ What an agent can be given · your call

Stripe Notion Linear Jira Figma HubSpot Shopify Google Sheets Your accounting tool Your CRM Any REST API MCP servers

/ Systems they touch · live

GitHub Gmail Vercel Supabase X Solana via Helius

/ Infrastructure

Supabase Node.js PostgreSQL Vercel

/ Control & safety

Human approval gates Role-scoped access Full audit log Spend limits Rollback

/ Frequently asked

Short and
honest.

What is Orchestra AI?

Nine AI agents that do the work instead of advising on it. You send one sentence in a chat. It goes to whichever agent fits, they carry it out, and it comes back finished. Kai opens the pull request. Clara writes the invoice or the contract as a PDF. Mia writes the post, Nina makes the image. Elena reads the actual sources and tells you what she could not verify.

What does it actually do for me?

It takes on work you would otherwise hire for. A developer opening pull requests, a designer, a researcher, someone answering the same community question for the fortieth time, someone writing your invoices. The work does not disappear, it just stops sitting on your desk.

Do the agents act on their own?

No. Nothing leaves the building without your yes. Not a post, not an email, not a merge, not an invoice. That is not a setting you can forget to switch on, it is how the system is built. Missing control, not missing intelligence, is what Gartner says kills most agent projects.

Where do they run?

Telegram and the chat on this page today. Discord, Slack, WhatsApp Business or your own app is an integration, not a rewrite. They answer in the language you write in, and you can simply send them a voice note instead of typing.

Can they work with our own systems?

Yes. Stripe, Notion, Linear, Jira, Figma, HubSpot, Shopify and your own APIs. What an agent is given is your decision, not ours. On request we build agents directly on your systems and teach your team to keep building them.

What does it cost?

Nothing at the moment. Orchestra AI is a pilot project and free of charge — there is no offer and no pricing. If that changes we will say so here and write to everyone on the waiting list.

How do I try it?

Not at the moment — the agents are offline while we rebuild. No account, no card, no obligation. Ideally not a test question, but something you have been putting off.