Banya LabsBanya Labs/Apps/Marii
In progress

App 01  ·  FIG. 001

Marii logo

Marii

AI-powered offline-first quotation & invoice engine for SMEs

SMEAIQuotation

Customer Target

Southern African SMEs, Trades & Contractors

Status

In progress

Agent Surface B

MCP Ready

Model Context Protocol Enabled

Registered in /.well-known/mcp.json & /llms.txt with deterministic Zod schema validation.

About Product

Marii is a ZAR-first quotation and business management engine tailored for Southern African SMEs and service providers. It parses voice notes, plain text prompts (English/isiZulu), and uploaded RFQ documents to automatically extract line items and auto-match with your Business Catalogue. Built on Next.js with Dexie IndexedDB offline storage, multi-tenant boundaries, PDF generation, and automated quote dispatch via persistent Evolution WhatsApp API instances.

Dual-Surface Execution

Human vs. Agent Flow

Every action in Marii is mapped bi-directionally across Surface A (Visual UI) and Surface B (Programmatic MCP Engine).

Surface A — Human Ergonomics Flow
  1. 01Upload RFQ document, record audio note, or enter plain-text requirements
  2. 02AI parses line items, calculates VAT, and matches against Business Catalogue
  3. 03Generate PDF quotation and dispatch directly to client via Evolution WhatsApp API
Surface B — Agentic Machine Execution Rails
Protocol: Model Context Protocol (MCP 1.0)Auth: Bearer banya_ag_...

1. Discovery: Scraper/Agent fetches /llms.txt or /.well-known/mcp.json manifest.

2. Tool Invocation: Executes deterministic Zod-validated tool contracts matching marii-quotes operations.

3. Replay Protection: Enforces Idempotency-Key: <uuid> to prevent duplicate mutations during LLM retry loops.

4. Token Compression: Returns structured, PII-minimized Markdown/JSON stripping all layout noise.