Cortex

A complete AI environment — chat, skills, memories, and code. Your main agent, or its helper.

Mobile-first app (Expo + .NET) that talks to multiple LLM providers (OpenRouter, Ollama — direct connectors coming) through its own backend with OAuth authentication (Google + GitHub), JWT, and SSE streaming. The goal: manage all your AI context — skills, memories, indexed code — on mobile and desktop, working as your main agent or as a helper to other agents. Offline is free for non-commercial use; cloud sync will be a subscription. See docs/ROADMAP.md for the full plan.

Architecture

flowchart LR
  subgraph Mobile["📱 Cortex (Expo)"]
    UI["Screens<br/>Login · List · Chat · Settings"]
    State["Zustand stores<br/>auth · conversations"]
    SSE["SSE client<br/>(streaming tokens)"]
  end

  subgraph Backend["🖥️ Cortex.Core (.NET 10)"]
    API["ASP.NET Core<br/>JWT · CORS · OpenAPI"]
    Auth["OAuth<br/>Google + GitHub"]
    Svc["Services<br/>Chat · Conversations · Models"]
    Providers["Providers<br/>OpenRouter · Ollama"]
  end

  subgraph External["🌐 Providers"]
    OR["OpenRouter API"]
    OL["Ollama (local)"]
    G["Google OAuth"]
    GH["GitHub OAuth"]
  end

  UI --> State
  UI --> SSE
  SSE -->|"POST /api/chat/{id}<br/>text/event-stream"| API
  State -->|"REST /api/..."| API
  API --> Auth
  API --> Svc
  Svc --> Providers
  Auth --> G
  Auth --> GH
  Providers --> OR
  Providers --> OL

Loading

Stack

• Backend: .NET 10, ASP.NET Core, EF Core 10, PostgreSQL 17, JWT, OpenAPI
• Frontend: Expo SDK 56, React 19, React Native 0.85, expo-router (typed routes), styled-components, Reanimated 4, Zustand, expo-secure-store, react-native-markdown-display, react-native-gesture-handler
• Infra: Docker Compose (Postgres), EAS Build (dev build)

Structure

Cortex/
├── Cortex.Core/              # .NET backend
│   ├── Controllers/          # Auth, Me, Conversations, Models, Chat
│   ├── Services/             # AuthService, ChatService, OAuthService...
│   ├── Providers/            # IProvider, OpenRouterProvider, OllamaProvider
│   ├── Data/                 # AppDbContext
│   ├── Objects/              # Domain entities
│   ├── Auth/                 # JWT options, CurrentUser
│   └── Dtos/                 # Request/response records
└── CodePlus/                 # Expo frontend (to be renamed)
    ├── app/                  # expo-router routes
    │   ├── (tabs)/           # Home + future tabs
    │   ├── conversation/     # [id] chat + new picker
    │   ├── login.tsx
    │   ├── callback.tsx
    │   └── settings.tsx
    ├── api/                  # HTTP client + SSE + typed endpoints
    ├── components/           # ui/ chat/ sheets/ feedback/ markdown/
    ├── stores/               # authStore, conversationsStore
    ├── theme/                # colors, motion, ThemeProvider
    └── config/               # env-driven runtime config

Design Decisions

• Backend as the single proxy: OpenRouter/Ollama keys never touch the client. The frontend is stateless.
• Server-side persistence: conversations/messages live in Postgres. The app keeps only the JWT + theme preference.
• SSE over POST: token streaming via text/event-stream, read in the app through fetch's native ReadableStream.
• Gray-first theme: grayscale palette with a white variant; consistent accent across modes.
• Intentional animations (Emil Kowalski philosophy): buttons with scale(0.97) on press, bubbles with staggered entrance (45ms), BottomSheet with damping + velocity-based dismiss, toasts entering/exiting in the same direction via interruptible transitions.

Roadmap

• Phase 1: Chat app (mobile + backend)
• Phase 2: Provider layer (model picker, direct connectors, BYOK, routing) ← we are here
• Phase 3: Context engine (memories, skills, code indexing, RAG, MCP)
• Phase 4: Agent mode (tools, orchestration, helper mode, automations, A2A)
• Phase 5: Desktop + Cloud (desktop app, E2E sync, subscription)

Full backlog with per-phase tasks: docs/ROADMAP.md.

License

See LICENSE.