An AI admin worker for agencies
Account managers at small agencies spend hours a month on work that follows rules: retainer maths, hours against budgets, the same summary for every client. That is what OpenClerq is for.
Short answer
OpenClerq takes the rule-based back-office work off an agency's account managers. Retainer overages and project margins run as formulas in a deterministic engine, a script can post each week's hours for checking and explanation, month-end summaries are drafted from the engine's figures, and every run is recorded with its cost. It runs on your own machine, with the model you choose.
The admin that eats an agency's month
- Working out retainer overages — hours logged against hours included — for every client, every month.
- Project margins and effective hourly rates, recalculated whenever a budget moves.
- Month-end summaries that say the same things in the same order for twenty clients.
- Checking weekly hours before they turn into a billing surprise.
- Remembering the Monday checks nobody owns.
None of it needs judgement until the end. All of it needs to be right.
What OpenClerq does for an agency
| Task | How | Status |
|---|---|---|
| Retainer overage and invoice totals | A formula spec in the calculation engine, with a proof record per run | Today |
| Project margin and effective hourly rate | Formulas; the model explains what moved and why | Today |
| Weekly hours check | Your time-tracker export script posts the week's figures to an OpenClerq webhook; the engine calculates, the model explains, the run is recorded | Today |
| Month-end client summary | Give it the month's figures and your summary template; the engine totals them and the model drafts the summary | Today |
| Monday admin checklist | A scheduled trigger produces it every week and records the run | Today |
| Picking the cheapest model that is good enough | Comparison mode: one brief, up to eight models, cost and time for each | Today |
| Reading the time-tracker export itself and flagging mismatches | The tool-calling loop, inside the policy engine and sandbox | 0.6 |
| A budget per automation | Pauses the automation instead of overspending | 0.6 |
| Connecting your time tracker, invoicing and CRM | A module built around your stack | Module |
Example: a retainer overage, calculated the same way every month
Write the rule once, as formulas with named inputs. Formulas run in the order they are listed, so later ones can use earlier results.
{
"spec": {
"formulas": {
"overage_hours": "max(logged_hours - included_hours, 0)",
"overage_fee": "overage_hours * overage_rate",
"invoice_total": "retainer_fee + overage_fee"
},
"output_names": ["overage_hours", "overage_fee", "invoice_total"]
},
"inputs": { "logged_hours": 46.5, "included_hours": 40,
"overage_rate": 95, "retainer_fee": 3800 }
}The engine returns an overage of 6.5 hours, an overage fee of 617.5 and an invoice total of 4,417.5, with a proof hash that ties those outputs to those inputs. A month under the included hours gives an overage of zero, not a negative number. The model then writes the line for the client: what was included, what was used, and why the invoice is higher this month.
The same approach covers project margin — round((billed - cost) / billed * 1000) / 10 gives a percentage to one decimal — and effective hourly rate, round(billed / hours * 100) / 100.
A week with OpenClerq
| When | Trigger | What happens |
|---|---|---|
| Monday 08:00 | 0 8 * * 1 | The week's admin checklist, recorded as a run |
| Friday 16:00 | Webhook from your export script | Hours against retainers calculated and explained, one run per client |
| First of the month | 0 9 1 * * | Month-end reminder with the list of summaries to prepare |
| Any time | Ask panel | "Why is the Northwind margin down?" — with the figures from the engine |
Every run appears in the run history with its steps, model and cost. See how to schedule recurring AI tasks for the setup.
What stays with people
- Pricing, scope and every conversation with a client.
- Approving and sending invoices — OpenClerq prepares the figures and the wording; it sends nothing.
- Deciding what an overrun means for the relationship.
Client data stays in the agency
Client budgets, rates and margins are commercially sensitive. With a local model through Ollama or LM Studio, nothing leaves the machine. With a cloud model under your own key, the context preview shows exactly what will be sent before it is. The gateway listens only on the computer it runs on and requires a token for every request. See how to use AI with confidential client data.
Frequently asked questions
Can OpenClerq send invoices or emails to clients?
No. It calculates, explains and drafts; a person reviews and sends. It has no email integration.
Does it connect to our time tracker?
Not natively. A small script can post the week's figures to an OpenClerq webhook, and OpenClerq can call HTTPS APIs on an allow-list you set. A module built around your time tracker, invoicing and CRM is something we build through Clerq for Business.
Can the whole team use it?
OpenClerq is a single-user desktop app, so each person runs their own. Shared, hosted deployments for teams are part of Clerq for Business.
What does it cost to run?
The software is free. With a local model there is no per-token cost; with a cloud model you pay your provider, and OpenClerq records the cost of every run so you can see what each automation costs.