BASELINE
Core SaaS baseline
A full-stack starter that treats tenancy, auth, permissions, billing, audit, and CI as foundation, not backlog.
Next.js App RouterFastAPI service layerPostgres migrationsJazzCash lifecycle
MVP FACTORY METHOD
Generate production SaaS systems with AI-ready architecture.
MVP Factory turns the repeated parts of a serious product into a reusable baseline: tenant-safe backend, composed frontend, billing rails, AI modules, tests, and deployment gates.
factory-previewready
Generation command
make create
01
Name the product
Capture display name, slug, output directory, and deployment defaults before any files are written.
02
Choose the architecture
Select Minimal, Standard, Full AI, or Enterprise, then choose multi-tenant or single-tenant generation.
03
Toggle capabilities
Enable billing, file uploads, notifications, invitations, RAG, agents, AI API, and workflow modules only when needed.
Next.js 16
app router interface layer
FastAPI
service-first backend spine
Postgres
Schema-per-tenant isolation
633
backend tests tracked as factory signal
Next.js 16
app router interface layer
FastAPI
service-first backend spine
Postgres
Schema-per-tenant isolation
633
backend tests tracked as factory signal
BASELINE SYSTEM
A factory baseline shaped like a product team already built it.
The reference site uses repeated calm blocks to explain a complex AI product. Here, those blocks explain the reusable architecture layers MVP Factory gives every generated app.
BOOTSTRAP
Generation workflow
The factory wizard turns a product name, preset, module mix, and tenancy model into a ready project structure.
Project presetsModule pruningTenancy modesDeployment defaults
INTELLIGENCE
AI platform layer
Provider-neutral inference, RAG, agents, workflows, evals, and observability plug into the baseline without leaking into app code.
Provider fallbackPrompt registryRAG ingestionAsync evaluation
GUARDRAILS
Security and correctness
Tenant-first search paths, permission gates, audit events, weak-secret checks, and CORS rules make safety a default behavior.
Tenant schemasRBAC checksAudit sinkConfig validation
AI PLATFORM LAYER
Intelligence without scattering AI concerns through the app.
Cosmoq presents automation as clean capability groups. MVP Factory maps that method to provider contracts, compliance gates, RAG, agents, workflows, and evaluation loops.
Provider-neutral inference
OpenAI, Anthropic, Gemini, and local providers sit behind one provider contract with retry and fallback.
Compliance pipeline
PII redaction, content safety, data residency, token budgets, and audit logging wrap every full AI call.
RAG and citations
Document ingestion, hybrid search, reranking, streaming chat, and source citations ship as a cohesive slice.
Agent runtime
A ReAct loop, tool registry, run traces, and long-term memory give generated products extensible automation.
Workflow DAGs
LLM calls, tools, code, and condition steps compose into retryable workflows with parallel branches.
Evaluation and cost visibility
Async evaluators, usage tracking, and per-model cost calculations keep AI behavior observable after launch.
OPTIONAL MODULES
Add only the rails your product actually needs.
Optional modules are designed to prune cleanly. A generated project that disables a capability should compile without stale references.
billing
Billing rail
JazzCash hosted checkout, signed return handling, usage metering, and subscription routes.
StandardFull AIEnterprise
file_uploads
Storage intake
Multipart upload endpoints, object storage adapters, quotas, and tenant-aware file records.
StandardFull AIEnterprise
notifications
Signal bus
Email and in-app notifications with templates, delivery workers, and user preferences.
StandardFull AIEnterprise
invitations
Clinic access
Invite-by-email flows, token validation, and clinic membership on accept.
Enterprise
ai_api
Inference API
Direct completions, streaming, and structured output endpoints behind the compliance pipeline.
Full AIEnterprise
rag
Knowledge fabric
Knowledge bases, ingestion loaders, hybrid search, reranking, and citation-ready retrieval.
Full AIEnterprise
agents
Agent runtime
ReAct execution, built-in tools, memory, and traceable agent runs for operational workflows.
Full AIEnterprise
workflows
Workflow engine
DAG-based LLM, tool, code, and condition steps with retries and parallel branches.
Full AIEnterprise
GENERATION WORKFLOW
From product prompt to generated architecture.
The factory wizard behaves like an architectural intake form: it captures the system shape, then emits a project that already follows the repository layer boundaries.
project generation4 steps
$ make create
Display name: Medical AI Scribe
Preset: Full AI
Modules: billing, rag, agents, workflows
Tenancy: multi-tenant
Validating module specs and migration tracks
Name the product
make createCapture display name, slug, output directory, and deployment defaults before any files are written.
Choose the architecture
Select Minimal, Standard, Full AI, or Enterprise, then choose multi-tenant or single-tenant generation.
Toggle capabilities
Enable billing, file uploads, notifications, invitations, RAG, agents, AI API, and workflow modules only when needed.
Validate the factory
python bootstrap/validate_spec.pyModule specs, prune rules, migrations, CI templates, and generated docs are checked before handoff.
SECURITY AND DELIVERY
The generated app inherits the safety checks first.
The reference security section maps naturally to the strongest MVP Factory promise: tenant correctness and deployment readiness are built into the baseline.
Tenant boundary
Schema-per-tenant isolation
Every tenant request sets `search_path` to the clinic schema first, then public, preventing table shadowing.
Mutation boundary
Service-owned permission checks
Routers validate and delegate; services own authorization, orchestration, and write-side audit behavior.
Configuration boundary
Runtime safety checks
Weak secrets, permissive production CORS, and unsafe RAG paths fail fast instead of becoming deployment surprises.
Recovery boundary
Audit events stay recoverable
Audit sinks are fire-and-forget, but failures are logged so user-impacting writes can be recovered from app logs.
TypeScript
0
Frontend type errors tolerated before build
Backend tests
633
Recorded test corpus across service, tenancy, AI, and bootstrap slices
Factory smoke
on
Generated project validation catches broken module pruning and stale templates
CI stages
6
Lint, test, smoke, build, staging migration, production migration gate
READY BASELINE
Start from a baseline that already knows the hard parts.
Use the factory as the first draft of a serious SaaS product, then spend your energy on the domain that makes it yours.
Sign in