How it works.

One week to map your company. Two weeks to make the first job live. Then we run it, on accounts you own, with a person approving what ships.

One week to map it. Two weeks to live. Then it runs, and we watch it.

Week 1 · the audit

Map

Kick-off Monday. We interview the people who do the work, connect read-only to your tools, and map every recurring job, handoff and spreadsheet. Friday: the report and a walk-through. $1,500, credited against your first install.

Days 1–10

Build

On accounts you own, in your own private setup. The routine steps run the same way every time; AI is used only where judgement is needed.

Day 14

Live

First scheduled run, approved by your team. We deliberately break one step to show it stops and asks instead of sending. Your team gets the one-page guide.

Every week

Operated

We watch every run, fix anything that breaks, check the quality, and send a monthly report of what it saved. One planned improvement a month.

Live and approved by day 14, or the install fee is refunded.

The install and the monthly are scoped and priced from your audit report, so you never pick a plan blind. Whatever happens, you keep everything.

  • Runs on accounts you own: your AI account, your tools, your spend limits
  • Everything it knows about your clients, and all your data, is handed to you if you leave
  • A person approves anything before it reaches a client

What a week looks like once it runs.

Three jobs, one Slack workspace, one person approving. Example week; names are fictional.

Reporting agentDrafted 10 client reports · 1 odd number held
approved
Brand QA agentChecked 14 captions · 3 fixes, each citing a rule
2 pending
Trends & ideas12 briefed ideas · 4 added to the calendar
filed

Your monthly report shows the real numbers: runs, hours saved, mistakes caught before a client saw them.

Your own private setup, nothing shared Connections you approve, no passwords A person approves before anything ships
See every integration1,000+ tools, across the whole company

Stay on the best model.

Your connections, approvals, client knowledge and monitoring sit with Compoundr, not inside any one AI model. When a better or cheaper one comes out, we switch. Nothing else changes.

  • The best model where judgement matters, the cheapest where it doesn't. Chosen per step.
  • AI usage runs on your own account with a cap you set. We set it up with you on day one; typical spend is £40–120 a month.
  • Try a new model on one client first, then roll it out. Same tools, same approvals, no rebuild.
Any leading AI model
ClaudeGPTGeminiOpen-source models
your Compoundr layer · connections · approvals · client memory · monitoring
SlackSheetsMetaInstagramNotionAsana
Your tools, connected once
Built to be safe

Acts on your behalf. Never puts you at risk.

Agencies hold other people's brands and data. Every job we run is limited to what it needs, signed off by a person, and kept apart from everyone else's. The agent can do the work without being able to do damage.

Guardrails

Only the access it needs

Every connection is limited to the job. A reporting agent can read insights and post a draft; it cannot delete a post, change a budget or send to a client. You choose what it can and can't do, per tool.

reporting agent
read insightsread mediapublishdelete
read insightsedit budgetcreate ad
post to #reportingDM clients
No passwords required

It never sees your passwords

You approve each connection yourself, so we never hold a password. There is nothing to leak, and no trick message can make the agent hand one over. Revoke in one click.

what we never store
Meta password
TikTok login
Slack login✓ your approval · limited to the job · switch off any time
Nobody sees your data

Kept apart, approved per action

Your agency gets its own private setup. Your clients' data never touches another agency's agent, is never used to train anything. GDPR-ready; data-processing agreement available.

Fair questions.

Could my team just do this with ChatGPT?

Most of it, honestly. What it won't do is stop itself quietly sending a client the wrong number, or tell you which brand rule a caption broke. That is what you pay for: kept separate, watched, checked, signed off, and stopped safely when something looks wrong.

Whose accounts does it run on?

Yours. Your AI account, your tools, your spend limits. The agent runs on its own private computer in the cloud that we manage; you authorise it, and you can revoke it in a click. We never hold your passwords.

What happens if we stop working together?

You keep everything. Everything the agent knows about your clients, and all your data, is handed to your own accounts when you leave. After the minimum term, thirty days' notice; nothing switches off.

Is our client data safe?

Each agency gets its own private setup and its own storage. You approve every connection yourself. Nothing is shared between our clients, and nothing goes out without a named person approving it. GDPR-ready; a data-processing agreement is available.

Which AI does it use, and which tools?

Whichever leading model is best for each step, on your own account. Tools: over a thousand, including the social platforms, ad accounts, sheets, drives, project tools and Slack most agencies already run. If a tool can talk to other software, it can almost certainly plug in.

How is the work after the audit priced?

From the report. Each job gets a fixed install price and a flat monthly for running it, based on what the audit found. Founding clients keep their founding rate. The audit fee is credited against the first install.

What do you need from us in the two weeks?

One named approver. One-click connections to the tools you already use. Your existing brand rules per client, in whatever form they exist. Two short calls, and about an hour of your team's time to review the first live run.

Book the audit$1,500 · one week · full automation map