The website was step one. The agents run the business.

JustCopy.ai builds a small business its website from a plain-language description. The next thing the owner does is launch agents to run that business: one that answers customers now, one that works in the background every day. JustCopy uses oncell as the runtime for both, one agent per business, metered per business, and getting better from its own conversations.

The customer

JustCopy sells to people who do not write code and do not want to: restaurant owners, salons, clinics, tradespeople, coaches, small online stores. They type what they want, watch a page appear, and publish it on their own domain. Three engines sit behind that: instant prebuilt apps chosen by a classifier, an archetype engine that fills designed templates in seconds, and a general coding agent for everything else. Plans run from $9.99 to $49.99 a month, priced by how many things a customer can ask for per day.

Two facts about this customer shape everything below. They will never see infrastructure, so the runtime has to disappear completely. And every one of them is a separate tenant with a separate budget, so cost has to be known per business, not per platform.

From “build my website” to “run my business”

Every JustCopy website already ends in a form: a table reservation, an appointment request, a free-trial booking. That form is a front desk rendered as HTML. The owner’s next request is the same description with a verb: “answer my customers and book them,” “follow up with everyone who asked for a quote.” JustCopy turns the description into an agent the way it turns one into a page. oncell is what makes that agent run, remember, wake up on a schedule, and stay accountable.

Two kinds of agents, one agent runtime

The flow, end to end

Sequence matters here: everything after step four happens on oncell, and JustCopy never provisions a thing.

1. The website exists. JustCopy built it from a description. Its booking form is the only front desk the business has.
2. Launch. On the project card the owner presses Launch an agent, picks "answers customers now" or "works in the background," and types one sentence about the job.
3. Build. JustCopy writes a brief from the website's own facts with one model call, then generates a one-file oncell agent: instructions, model, a chat handler, and for background agents a run task plus a daily schedule.
4. Deploy. One call to the deploy endpoint under a name that carries the customer id, with a $1 per day budget. oncell stores version one and assigns the cell. Nothing else is provisioned.
5. Run, sync. A customer's question goes to the chat task. The cell wakes, the model call goes through oncell's gateway where it is metered and logged, and the reply comes back in about a second with its cost and run id.
6. Run, async. The daily schedule is a durable wake owned by the runtime. The agent parks at no cost between wakes. Run now invokes the same task by hand.
7. Watch. JustCopy reads run history per agent and shows the owner recent conversations and runs. Every run is a full trace on oncell, so a disputed reply is a run id away.
8. Evaluate. A bad run, like the plumber's invented price below, is promoted into that agent's evals. A template change ships only after the eval set passes.
9. Version and roll back. Every template improvement redeploys the customer's agent as a new version. If the evals regress, rollback is one call, and JustCopy never touches the customer's cell.
10. Improve. The same evals on two models decide the cheapest one that still passes. The run log is the trajectory set for a small model of JustCopy's own later.

Every oncell feature, and what JustCopy does with it

Live means exercised against api.oncell.ai from JustCopy’s code on September 4, 2026. Next means the oncell primitive exists and the JustCopy side is on the list below.

Three businesses, live on oncell today

These are real agents deployed under JustCopy’s oncell account during this study, one per business, each from a description a JustCopy customer would actually type. The replies below are verbatim, with what each one cost.

Port Orchard Bakery · front desk

Cascade Dental · scheduling

Summit Plumbing · follow-up

Summit Plumbing’s evals

6 of 7 passing on claude-haiku, average $0.0007 per reply, 1.8 s. The one failing case is a tone rubric judged by a model; it is the kind of regression a business owner never sees but a platform should. How promotion works is in the docs under Promoting a production run.

The per-customer cost of AI agents: what JustCopy gets per business

How it plugs in

JustCopy already has most of this. Its restaurant, clinic, and studio templates all end in a booking form, which is a front desk rendered as static HTML. It already has a per-user agents table and a working agent chat page. What was missing was a runtime. The integration is one backend route file, a Launch an agent button on every website project, and one agent page: a chat thread for agents that answer customers, a Run now button and run history for agents that work in the background. No new infrastructure, no per-customer keys: JustCopy’s one oncell key deploys every agent under a name that carries the customer id, so the tenant boundary is the name.

The owner says what the agent should handle; JustCopy fills in the business from the website it already built and turns that into the agent’s instructions, exactly the way its archetype engine turns a description into a page today. A background agent gets one extra line in its source, a daily schedule that oncell owns. Then the calls:

POST https://api.oncell.ai/api/v1/deploy
Authorization: Bearer $ONCELL_API_KEY
{ "agentName": "jc-<customer>-front-desk",
  "source": "<generated agent file>",
  "manifest": { "identity": { "instructions": "...", "model": "claude-haiku",
                              "budgets": { "perDayCents": 100 } } } }

POST https://api.oncell.ai/api/v1/agents/jc-<customer>-front-desk/chat
{ "message": "are you open monday morning?" }
→ { "text": "We're closed Sundays and Mondays...", "cost": 0.00041, "runId": "run-..." }

POST https://api.oncell.ai/api/v1/agents/jc-<customer>-follow-up/run
{ "prompt": "Do today's follow-ups." }        # or let the daily schedule fire
→ { "text": "3 customers asked about cakes yesterday and did not order. Drafts: ...", "cost": 0.0012 }

The daily budget is set per customer at deploy time and enforced below the model, so a runaway conversation on a $9.99 plan stops at $1 a day without JustCopy writing a line of billing code. The mechanism is described under Budgets; the daily wake under Schedules.

What we are doing next together

1. Ship the route and the button in JustCopy, behind a flag, for ten bakery and salon customers.
2. Give every JustCopy template a starter eval set (hours, prices, tone, cost) so new agents are guarded from their first conversation.
3. Route each business's agent to the cheapest model that still passes its evals; measure the margin.
4. Post-train a small model on JustCopy's trajectories when there are enough conversations, and serve it through the same gateway.

All figures in this story are from live runs on oncell on September 4, 2026, under JustCopy’s founder account. JustCopy user and project counts are JustCopy’s own figures.