A visually charming, personality-driven AI companion built with privacy, user agency, and emotional safety as core principles.

Try it live: https://athena-seven-beta.vercel.app

• Multi-turn conversation with encrypted persistent memory (IndexedDB, AES-GCM, PBKDF2 600k iterations)
• Configurable memory window (1–10 messages)
• Document & image attachments (txt, md, json, csv, pdf, images, code files)
• Web search — OpenAI via Responses API, Groq via Compound, OpenRouter via openrouter:web_search
• Image generation — inline via OpenAI image_generation tool with download overlay
• Emotion detection — post-response classification displayed as an emoji badge
• Token usage display per message
• Markdown rendering with syntax highlighting and emoji support

• STT: Groq / OpenAI Whisper — tap-to-record with waveform visualizer
• TTS: OpenAI TTS (6 voices) and ResembleAI Chatterbox
• Per-message audio replay
• TTS never reads URLs aloud

• All data encrypted locally — AES-GCM + PBKDF2 600k iterations
• Export integrity: SHA-256 HMAC in every export file
• Privacy mode — suppresses Vercel Analytics entirely
• OpenAI store: false — conversations never persisted on OpenAI servers
• No lock-in — export as JSON or Markdown, delete anytime

• Installable on desktop and mobile
• Service worker + web app manifest
• Offline-capable shell

• Node.js 18+
• At least one API key — Groq is the easiest free starting point

git clone https://github.com/Jiab77/athena.git
cd athena
npm install
npm run dev
# Open http://localhost:3000

1. Open the app at http://localhost:3000
2. Click the FAB button (bottom right) → Settings
3. Go to Model → select Groq → paste your API key
4. Select a model → start chatting

All keys are configured inside the app under Settings > Model — encrypted with AES-GCM before being stored in IndexedDB, never sent anywhere unencrypted.

1. Click the FAB button → Settings
2. Go to Companion — choose from 30+ presets (category, gender, color scheme) and set a name
3. Go to Customize — pick a personality or define a custom one, then choose the visual format
4. Save — stored locally, encrypted

See docs/ROADMAP.md for details.

/
├── app/ # Next.js App Router routes
├── components/ # React components (chat, avatar, settings, UI)
├── hooks/ # Custom React hooks
├── lib/
│ ├── llm/ # LLM providers (Groq, OpenAI, BioLLM, Custom, router)
│ ├── voice/ # TTS providers (OpenAI, ResembleAI)
│ ├── db.ts # IndexedDB schema + encrypted settings
│ ├── crypto.ts # AES-GCM encryption (PBKDF2 600k iterations)
│ ├── export.ts # Export pipeline + SHA-256 integrity
│ └── import.ts # Import pipeline + validation
├── public/
│ ├── avatars/ # 30 cyberpunk character avatars
│ ├── images/ # Screenshots and backgrounds
│ ├── sw.js # Service worker
│ └── manifest.json # PWA manifest
└── docs/ # Architecture, roadmap, security audit

Design Inspiration: GateBox | Project AVA | VirtuaGirl

LLM Providers: Groq | OpenAI | OpenRouter | BioLLM | Ollama

Voice: OpenAI TTS | ResembleAI Chatterbox | Groq Whisper

Avatar: Decart AI | React Three Fiber | Three.js

UI & Storage: shadcn/ui | Tailwind CSS v4 | IndexedDB

MIT License — see LICENSE for details.

Single-developer project with AI agent collaboration. Ideas are welcome — open an issue to discuss before submitting a PR.

This project was built using a custom AI collaboration framework designed to maintain session continuity, project context, and behavioral consistency across all development sessions.

Built with Jiab77/ai-framework — a lightweight, project-agnostic framework for working effectively with AI models during development. Also used in the Virgil project.