5.1 Scenario: Building "Bento" — a Meal Prep SaaS
Use Case: Building "Bento" — a Meal Prep SaaS
Let's walk through how the ALAI system would build a completely new product from CEO idea to deployed MVP.
The Idea
Alem: "We need a meal prep subscription SaaS for the Nordic market. Next.js frontend, Kotlin/Ktor backend, PostgreSQL. Call it Bento."
Step 1: Task Creation in Mission Control
Alem (or John) creates the initial tasks:
node mc.js add "Bento: Project scaffold — Next.js frontend + Kotlin/Ktor API" -p H -t alem
node mc.js add "Bento: Database schema design — users, subscriptions, meals, deliveries" -p H
node mc.js add "Bento: Landing page design — Nordic minimalist, hero + pricing + CTA" -p H
node mc.js add "Bento: API — user auth + subscription management endpoints" -p H
node mc.js add "Bento: Frontend — meal selection + checkout flow" -p H
These 5 tasks enter MC as [open] status.
Step 2: Pi-Orchestrator Picks Up Tasks
Within 30 seconds, the orchestrator wakes up:
[INFO] Found task #4001: Bento: Project scaffold — Next.js frontend + Kotlin/Ktor API
[INFO] Task #4001 classified: complexity=3 domain=code type=scaffold
[INFO] Task #4001 → llama-tool (kimi-k2.5:tq1 on forge)
Classification logic:
- "Project scaffold" + "Next.js" + "Kotlin/Ktor" → domain: code, type: scaffold, complexity: 3
- Complexity 3 → Tier 3 model (Kimi K2.5 on FORGE)
Routing:
- "Next.js frontend" → keyword match → CodeCraft (fullstack covers both)
- CodeCraft's soul + blueprints are injected into the prompt
Step 3: CodeCraft Builds It
The agent receives a prompt containing:
- CodeCraft's company identity and coding standards
- The
nextjs-app.yamlblueprint template - The task description
- Relevant HiveMind context (if any previous Bento-related work exists)
Agent output:
- Project structure created
package.json,docker-compose.yml,README.md- Basic Next.js pages + Kotlin API skeleton
- GOTCHA file written as proof of work
[INFO] Auto QA prep completed for task #4001
[INFO] Task #4001 completed via llama-server/kimi-k2.5:tq1 (912s)