Best AI Assistant for JobTread

JobTread already knows. Nothing tells you.

JobTread holds the whole job: the customer, the estimate, the schedule, the daily logs, the invoice, the payment. For a residential contractor it is usually the only system that has the complete picture.

The problem is that having the picture and acting on it are different things. An invoice goes out and the follow-up sequence exists only in someone's head. A payment lands and the next draw does not get scheduled. A daily log records a two-day weather delay and the three people who need to reshuffle around it find out at the Monday meeting.

That is where contractor cash flow actually leaks, and none of it is a JobTread problem. It is the space between what JobTread records and what somebody does about it.

The 42 events are the reason this works

JobTread publishes 42 webhook event types across accounts, contacts, jobs, locations, documents, document payments and recipients, payments, daily logs, tasks, time entries, comments, files, and form submissions. Crucially, they are self-serve and provisionable through the API itself, so Carly registers its own subscriptions instead of walking you through a settings screen.

Two of them carry most of the commercial weight. `documentSent` fires when an estimate or invoice actually goes out. The payment events fire when money moves. Invoice sent and payment received are the two moments where follow-up either happens or quietly does not, and both are directly triggerable.

One caveat worth knowing before you build: JobTread's documentation describes deliveries only as a detailed POST request. The payload shape, whether deliveries are signed, and the retry policy are not documented, so confirm signature verification against a live account rather than assuming it.

How Carly connects to JobTread

JobTread is not one of Carly's native connectors, so you connect it with your own credentials. JobTread publishes an open API called Pave, with documented authentication, OAuth, queries, mutations, file uploads, rate limits, and webhooks.

On carlyassistant.com/integrations, open Connect an API and paste JobTread's documentation link. Carly inspects the API, determines the authentication method deterministically, and asks only for the matching credentials. Carly then verifies that the API base answers before treating the connection as usable, so a connection that says connected has actually made a call.

Your credentials are encrypted at rest and injected only at execution time. Write access is a setting on the connection, and you can narrow it to specific paths instead of authorizing the whole API.

One shape note: Pave is GraphQL-like rather than GraphQL. It takes a nested JSON query object over a single endpoint, which behaves similarly but is not interchangeable with a GraphQL client.

What Carly can write in JobTread

The write surface is broad, which is what lets Carly finish a workflow instead of stopping at a notification. Create, update, and delete are supported across accounts covering both customers and vendors, contacts, jobs, locations, documents, document payments and recipients, daily logs, tasks, time entries, files, forms and submissions, cost codes and items, custom fields, payments, plans, roles, memberships, users, dashboards, webhooks, and workflows.

Three specifics will shape what you ask for. Estimates and invoices are the same object: there is no createInvoice, you create a document with a type such as customerInvoice, customerOrder, vendorOrder, vendorBill, or bidRequest. Creating a job has a mandatory chain, because a location is required to create a job and a customer is required to create a location, so a new job for a new customer is three sequential writes. And sendDocument genuinely emails the document to a recipient, which is real send capability rather than draft-only.

File upload is a two-step flow: create an upload request, PUT to the returned URL, then create the file.

JobTread also ships its own workflow engine and, since April 2026, an AI Connector built on MCP that works with Claude or ChatGPT at no additional cost. That connector is genuinely useful for asking questions about your job data. What it does not do is act on its own, which is the gap the 42 webhooks fill. The two are complementary rather than redundant.