Book a demo

The orchestration layer for AI agents

Many agents,one clockwork.

Orrery runs teams of specialised AI agents the way an orrery moves its planets. Each one keeps its own orbit, and all of them stay in time.

Book a demoHow it works
A brass orrery photographed from above, one planet painted ultramarine

Scroll, or press →

02 — The problem

A single brass sphere standing alone on an empty sheet of paper

One agent isn’t a team.

  1. 01

    It loses the thread.

    Long jobs outgrow one context window. Halfway through, the plan is gone.

  2. 02

    It does everything at one depth.

    Research, writing and review all get the same shallow pass.

  3. 03

    Nobody checks its work.

    A single agent marks its own homework, and errors ship quietly.

03 — The orrery

Specialists on their own orbits, kept in time.

  1. The conductor

    Reads the brief, splits the work and keeps time.

  2. Researchers

    Close orbit. They gather context and check facts.

  3. Builders

    Wide orbit. They write, code and produce.

  4. Shared memory

    Every agent reads from and writes to one record.

  5. The hand-off

    Work moves between agents, never dropped, always logged.

04 — How it works

Three steps from brief to finished work.

  1. 1

    Describe the work

    Write the outcome you want in plain language: the inputs, the bar for done, who signs off.

  2. 2

    Orrery assembles the team

    The conductor picks specialists, sets their orbits and wires them to one shared memory.

  3. 3

    Review, approve, ship

    Every hand-off is logged. You approve the result, or send it back around with a note.

05 — Use cases

Where teams put it to work.

  • Index cards connected by fine ultramarine ink lines

    Customer operations

    Triage, research and reply agents clear the queue overnight. People handle the conversations that need a person.

  • A brass magnifying lens over a printed star chart

    Research and analysis

    One agent gathers sources, another checks every claim, a third writes the brief your team actually reads.

  • Brass clock gears on gridded drafting paper

    Software engineering

    Planning, coding and review agents work the same ticket in turn, and every change arrives with its reasoning.

06 — Results

Measured in work that ships.

4.2×

more tasks finished per week than single-agent setups

63%

less time spent reviewing agent output

99.6%

of hand-offs logged, replayable and attributable

Median across 140 teams in the 2026 pilot programme

07 — Customers

What teams say

“We stopped babysitting one agent and started managing a team. Our overnight queue is empty by 7 a.m.”

Mara IlungaHead of Support Operations, Northwind Freight

“Every change arrives with the reasoning attached. Code review got faster because there's less to argue about.”

Tomás ReyesVP Engineering, Halden

“The fact-checking agent catches what our writers miss. It's the colleague who reads the footnotes.”

Priya RamanResearch Lead, Calder Health

08 — Pricing

Priced by the team, not the token.

  • Starter

    $49 per month

    For one team trying its first multi-agent workflow.

    • Up to 3 agents
    • One shared memory
    • 7-day hand-off history
    Start with Starter

  • Team

    Most teams

    $390 per month

    For teams running agents on real work, every day.

    • Up to 25 agents
    • Approval gates
    • Full audit log
    • Priority support
    Start with Team

  • Enterprise

    Custom annual agreement

    For organisations with security and residency requirements.

    • Unlimited agents
    • SSO and SCIM
    • Private deployment
    • Named engineer
    Talk to us

09 — Security

Built for teams that answer to auditors.

  • SOC 2 Type II

    Audited annually. Reports available under NDA.

  • Your region, your data

    Run in the EU, US or APAC. Nothing leaves it.

  • Every hand-off logged

    Who did what, with which inputs, and when. Replayable.

  • Human approval gates

    Agents pause for sign-off on anything you mark sensitive.

  • No training on your work

    Your prompts and outputs are never used to train models.

  • SSO and SCIM

    Provision and revoke access from your identity provider.

10 — Book a demo

Put your agents in orbit.

A 30-minute call with an engineer. Bring one real workflow and we'll set up its first team of agents with you.

01 / 10