Platform7 areas94 tables

Everything the business runs on, in one database.

Most operations stacks are a project tool, a CRM, a quoting spreadsheet, a mailing platform and a few monitoring subscriptions that have never met. YumaOS is the same work with nothing between the parts.

The full shipped inventory

Product snapshot

Why one database is the whole argument.

An agent asked what a job actually cost has to reach the estimate, the imported hours, the rate card and the invoice. If those live in four products, the honest answer is that it cannot.

94 tables

One application database holds the operational records. Agent memory sits separately on purpose, so applying the application schema can never take chat history with it.

Compound reads

A project comes back with its milestones, tasks, quotes and time actuals together, rather than as four round trips an agent has to stitch by hand.

42 screens

Twenty product screens and twenty-two settings screens, reachable from a keyboard-driven command palette, with server-sent events telling the page when something it is showing has changed.

One vocabulary

Task labels, expense categories and the other controlled lists are extended by an admin in Settings. Changing what the business calls something is not a deployment.

Five clusters, seven areas

The seven areas, and what is actually in them.

Every line below is a screen or a capability that exists today, not a roadmap item.

Work management

The board a delivery team opens every morning, and the register nobody wants to maintain but everyone needs.

Task board with configurable columns, drag ordering, labels, comments and @mentions
Projects with milestones and members, read back with their tasks, quotes and time actuals in a single call
Risk register with treatments, and an agent whose only job is keeping it current
Calendar in month, week, day and list views, and an editable org chart
A scheduled sweep that notices tasks, risks and quotes coming due before anyone does

Customers and commercial

Private working material on one side of a line, and the thing a customer is allowed to see on the other.

Customer register with contacts, addresses and lead, prospect, customer or inactive status
CSV import from a downloadable template. One bad row rejects the file rather than half-importing it
Estimating with cost rates, employer on-costs, subcontractor buy prices and gross margin, plus variance against actuals
Promotion from a private estimate to a draft quote, which is always a human decision
Quotes with line items, labour, materials and project context, exportable to Xero as a quote or an invoice

Money in and money out

The numbers match the ledger, because most of them are the ledger.

Invoices and payments mirrored from Xero, with an ageing view of who owes what
Expenses logged against categories and jobs. An agent may record spend and may never authorise it
Hours imported from whatever tracker the business already runs, by CSV, push endpoint or scheduled pull
An identity map that resolves the tracker's names onto real people, and a rate card that turns hours into job cost
Agent spend broken down by agent, model and channel, under a cap the agents cannot raise

Documents

Written material the business can search, and drafts that keep their history.

A document library chunked and embedded for retrieval, searchable from the UI or by an agent with citations
Document Studio drafts from templates: tender response, capability statement, risk report, one-pager or project status
Export to PDF, DOCX or HTML
Version history on every save, so a rewritten section is a new version rather than a loss
File storage on local disk, S3, Cloudflare R2 or Azure Blob, chosen by one setting

Reaching customers

Outward-facing work, with the send step held back on purpose.

Email campaigns with templates, segments, recipient lists and scheduling
Open and click tracking surfaced as engagement analytics
Signed one-click unsubscribe, a suppression list and consent records. A do-not-contact check that cannot be answered reads as suppressed
SMS through Twilio, AWS End User Messaging or Azure Communication Services, with inbound replies
Social posts drafted and reviewed in the product, published through Buffer once a person has approved them

Watching the market

Three scheduled sweeps, none of which calls a language model.

Competitor monitors that answer whether a page changed, record what changed and chart a price history
Guided setup that fetches one page and shows what it would extract before anything is saved
Tender and grant feeds scored against deterministic rules and a business profile, on a triage board
An on-demand fit assessment, which is the one model opinion here and is recorded with its reasoning
Media watches over Google News and custom feeds, with relevance rules and a triage step

Running the deployment

The screens that tell you the quiet week was actually quiet.

An audit log of every write, who made it, and on whose behalf, agent writes included
A schedules screen listing every cron, whether it is firing and what is orphaned. It deliberately has no agent tool
An organisation pulse digest, re-filtered against each recipient's own permissions at send time
Controlled lists an admin extends without a migration, and white-label branding applied on screen and in email
One deployment timezone, always an IANA identifier, read by everything that states a date

The agents use these screens, not a second copy of them.

There is no separate agent database and no synchronisation job to fall behind. A tool call reads and writes the same rows the screen does, under the same permissions as the person who asked.

How the agents are gated