Admin
Chart of accounts
The chart of accounts is empty. Import the starter chart (a draft Yardi-style tree, standing in until Brasa's real chart export arrives), then refine it here.
Custom KPIs
Add KPINo KPIs defined yet. Create the first one with Add KPI; every KPI is built from constrained blocks (metric, window, annualization, over a basis or loan balance), so each one is auditable by reading its row.
Reporting reminders
The engine checks every deal's expected core files (trial balance and rent roll) after its due day and emails the deal's configured recipients from the reporting mailbox. In development the send goes to the server console; the log below is the record either way. In production a daily schedule runs the same check automatically.
Recent sends (latest 0)
No reminders have been sent yet.
AI
The two AI features (loan document extraction, request template matching) fire only on explicit clicks, put everything through human review before any write, and are metered below. The app is fully functional with AI off.
The model, prices, cap, and trimming limits are the ai_* rows in Settings below. The API key lives in the server environment, never here.
Usage log (latest 0)
No AI calls have been made.
Settings
Firm-wide configurable values. Nothing here is hardcoded in the app; reports read these at computation time.
Loan document extraction sends at most this many PDF pages. Larger files are refused before any tokens are spent.
The request matcher sends the counterparty file's headers plus at most this many sample data rows, never the full data.
The Anthropic model string every AI call uses (BUILD_SPEC Section 13: cheap tier). If you change it, update ai_price_input_per_mtok and ai_price_output_per_mtok to that model's published rates.
Optional monthly AI spend cap in dollars. The word none means no cap. With a cap set, calls are refused once the month's estimated spend reaches it (or when spend cannot be computed).
Dollars per million input tokens for the configured AI model. Used to estimate each call's cost at call time.
Dollars per million output tokens for the configured AI model. Used to estimate each call's cost at call time.
Warn cushion for covenant tests, as a fraction of the threshold. 0.05 means a result within 5% of the threshold on the passing side shows Warn.
Body for reporting reminders. Placeholders filled at send time: {deal_name}, {period}, {due_day}, {reminder_number}, {missing_list}.
Days between repeat reminders for a deal whose reporting settings leave cadence blank. Whole number, at least 1.
Subject line for reporting reminders. Placeholders filled at send time: {deal_name}, {period}, {due_day}, {reminder_number}, {missing_list}.
Per-account, per-month tolerance in dollars for T12 validation against the trial balance. Differences at or below this are ignored.