Architecture
Platform Architecture
Full-stack TypeScript built on Next.js 16, Supabase, and Inngest. A 7-tier AI model ladder powers every interaction.
Foundation
Technology stack
Next.js 16
App Router with 845+ routes. Proxy-based authentication. Server and client components.
Supabase
12 schemas with Row-Level Security. Real-time subscriptions. S3-compatible storage.
Inngest
348+ background functions. Event-driven architecture with step functions and automatic retry.
TypeScript
1.2M lines. Zero any types. Strict mode with Zod validation at every boundary.
AI Models
7-tier intelligence ladder
63 models across 12 providers. Each tier has a fallback chain that ensures every request succeeds even during provider outages.
| Tier | Model | Cost (in/out per MTok) | Use Cases |
|---|---|---|---|
| Premium | Claude Opus 4.6 | $5.00 / $25.00 | Hardest problems, 1M context |
| Strong | Claude Sonnet 4.6 | $3.00 / $15.00 | Complex reasoning, nuanced synthesis |
| Standard | Gemini 3.1 Pro | $2.00 / $12.00 | Deep analysis, strategic planning |
| Light | Claude Haiku 4.5 | $1.00 / $5.00 | Chat, classification, streaming |
| Micro | Gemini 3 Flash | $0.50 / $3.00 | High-volume extraction, 1M context |
| Nano | Gemini 2.5 Flash Lite | $0.28 / $0.42 | Background analysis, proactive signals |
Fallback chains
Standard: Gemini 3.1 Pro, Gemini 3 Flash, Gemini 2.5 Flash, Claude Haiku
Premium: Claude Opus 4.6, Claude Sonnet 4.6
Runtime overrides
AI_MODEL_MICRO=gemini-2.5-flash
DISABLE_GEMINI=true
Data flow
End-to-end pipeline
Ingestion through entity resolution into the intelligence layer. Output surfaces through chat, voice, and proactive cards.
Ingestion Resolution Intelligence
────────── ────────── ────────────
Gmail Alias Lookup Workspace Brain
Outlook Semantic Match Living Memory
Slack ────────> Fuzzy + AI ────────> User Skills
Telegram Entity Creation Signal Router
WhatsApp
iMessage Scoring Output
Meetings ────────── ────────────
Webhooks Fellegi-Sunter Chat Pipeline
384-dim vectors Voice Interface
0.85 auto-merge Vibe Feed
MCP ServerDatabase
12 domain schemas
Schema-level domain separation with typed query helpers that enforce workspace scoping.
| Schema | Purpose | Key Tables |
|---|---|---|
| core | Users, workspaces, OAuth, membership | users, workspaces, oauth_tokens |
| crm | Contact management, entity resolution | crm_contacts, entity_identities, entity_aliases |
| intelligence | Meetings, analysis, vibe feed | meetings, meeting_sources, vibe_feed_cards |
| integrations | Channel messages across platforms | user_gmail_messages, user_slack_messages |
| tasks | Task management and projects | tasks, task_projects, task_categories |
| canvas | Document editor | canvas_documents, canvas_collections |
| admin | Usage logging, alerts | admin_ai_usage_log, admin_alerts |
| chat | Conversation persistence | chat_sessions, chat_messages |
| onboarding | Onboarding flow | onboarding_state, onboarding_steps |
| bookings | Scheduling, calendar | booking_links, booking_events |
0Files
0+Routes
0+Background Jobs
0Schemas
0Models
Talk to Sentigen
Have questions about what you're reading? The intelligence behind this platform can walk you through it.