Marlo

Agencies draft AI plans that never land on a board people trust, boards ignore real capacity, and many client accounts must stay isolated. Prompts also need to be versioned like product artifacts, not buried in config.

Shared infrastructure for speed, so every new feature must respect tenant boundaries and metering.

• A separate database per client, too expensive to operate early on.
• Prompts buried in env/config, no way for the team to version and improve them in-product.
• Plans in one tool and capacity in another, so risk never shows up next to the work.

Next.js App Router product for admin and client portals on one codebase. Plans are drafted with the Vercel AI SDK, then live on the same capacity-aware Kanban with at-risk flags. Prompt studio versions prompts in-product. Clerk handles RBAC; Drizzle + Neon Postgres for data; Stripe for usage-metered billing; OAuth to Google Analytics, Ads, and HubSpot. Covered with Vitest and Playwright.

• Prompts are versioned product artifacts, not one-off strings.
• AI plans sit next to capacity and risk on the same board.
• Admin and client experiences share one multi-tenant codebase with org-scoped isolation.

• Admin and client portals with metered Stripe billing.
• 30/90-day AI plans that land on a capacity-aware board.
• Connected planning via Analytics, Ads, and HubSpot OAuth.