Kai
An open-source AI assistant with persistent memory that runs on Android, iOS, Windows, Mac, Linux, and Web.
Overview
Kai is built with Kotlin Multiplatform and Compose Multiplatform. It connects to 30 LLM providers (plus a keyless Free tier and on-device models) with automatic fallback, remembers important details across conversations, and can act autonomously via scheduled heartbeats and tool execution.
Key Features
- Persistent memory — Kai remembers important details across conversations and uses them automatically
- Customizable soul — Define the AI's personality and behavior with an editable system prompt
- Multi-service fallback — Configure multiple providers; Kai automatically tries the next one on failure
- On-device inference — Run models locally with LiteRT on Android, iOS and desktop, no internet needed
- Tool execution — Web search, notifications, calendar events, shell commands, and more
- MCP server support — Connect to remote tool servers via the Model Context Protocol
- Interactive UI — The AI can answer with interactive screens (forms, buttons, multi-step flows) instead of plain text
- Skills — Install reusable instruction bundles in the SKILL.md format (Android)
- Autonomous heartbeat & scheduled tasks — Periodic self-checks plus one-time or recurring tasks the AI schedules
- Linux sandbox — On Android, the AI can run shell commands, scripts and packages in a Debian or Alpine environment
- Kai Build — An Android coding environment with terminals and coding agents (Claude Code, Grok, OpenCode)
- Encrypted storage — Conversations are stored locally with encryption
- Settings export/import — Back up and restore all settings as a JSON file
- Text to speech — Listen to AI responses
- Image attachments — Attach images to any conversation
How It Works
┌────────┐
│ User │
└───┬────┘
│ message
▼
┌─────────────────────────┐
│ Chat │
│ │
│ prompt + memories │
│ │ │
│ ▼ │
│ ┌────────┐ │
│ │ AI │◀─┐ │
│ └───┬────┘ │ │
│ │ tool calls │
│ │ & results │
│ ▼ │ │
│ ┌────────┐ │ │
│ │ Tools │─┘ │
│ └───┬────┘ │
│ │ │
└────────┼───────────────┘
│ store / recall
▼
┌─────────────────┐ hitCount >= 5
│ Memory │───────────────────┐
│ │ │
│ facts, prefs, │ ▼
│ learnings │ ┌────────────────┐
│ │◀─delete──│ Promote into │
└─────────────────┘ │ System Prompt │
▲ └────────────────┘
│ reviews
│
┌─────────────────┐
│ Heartbeat │
│ │
│ autonomous │
│ self-check │
│ every 30 min │
│ (8am–10pm) │
│ │
│ all good? │
│ → stays silent │
│ needs action? │
│ → notifies user│
└─────────────────┘
Supported Services
| Service | API Type |
|---|---|
| Atlas Cloud | OpenAI-compatible |
| Anthropic | Anthropic native |
| OpenAI | OpenAI-compatible (Responses API for models that require it) |
| Gemini | Gemini native |
| DeepSeek | OpenAI-compatible |
| Mistral | OpenAI-compatible |
| xAI | OpenAI-compatible |
| OpenRouter | OpenAI-compatible |
| Requesty | OpenAI-compatible |
| Groq | OpenAI-compatible |
| NVIDIA | OpenAI-compatible |
| Cerebras | OpenAI-compatible |
| Ollama Cloud | OpenAI-compatible |
| LongCat | OpenAI-compatible |
| Together AI | OpenAI-compatible |
| Hugging Face | OpenAI-compatible |
| Venice AI | OpenAI-compatible |
| Moonshot AI | OpenAI-compatible |
| Z.AI (standard and Coding Plan) | OpenAI-compatible |
| MiniMax | OpenAI-compatible |
| AIHubMix | OpenAI-compatible |
| Deep Infra | OpenAI-compatible |
| Fireworks AI | OpenAI-compatible |
| OpenCode | OpenAI-compatible |
| Public AI | OpenAI-compatible |
| AI Horde | OpenAI-compatible |
| 1min.AI | OpenAI-compatible |
| Perplexity | OpenAI-compatible |
| OpenAI-Compatible API (Ollama, LM Studio, etc.) | OpenAI-compatible |
| Local Model (Android, iOS, desktop) | LiteRT on-device |
Plus a built-in Free tier that requires no API key.
Platforms
| Platform | Distribution |
|---|---|
| Android | Google Play, F-Droid, APK |
| iOS | App Store |
| macOS | Homebrew, DMG |
| Windows | Winget, MSI |
| Linux | DEB, RPM, AppImage, Flatpak, tarball, AUR |
| Web | Browser |
Feature Documentation
- Chat & Conversations — Message history, conversation persistence, image attachments, and speech output
- Multi-Service — Provider configuration, fallback chain, and connection validation
- Tools — Available tools, execution flow, safety guards, and enablement
- Memories — Memory lifecycle, categories, reinforcement, and promotion
- Heartbeat — Autonomous self-checks, active hours, and configuration
- Tasks — Scheduled tasks, future execution, and task management
- Daemon — Background service for scheduled tasks and heartbeat execution
- MCP Servers — Remote tool servers over the Model Context Protocol
- Dynamic UI — AI-generated interactive screens rendered inline in chat
- On-Device Inference — Local LiteRT models, downloads, and integrity checks
- Reasoning — How reasoning traces are sent back to providers and shown
- System Prompts — How each prompt variant is composed
- Skills — Installable SKILL.md instruction bundles
- Linux Sandbox — On-device Debian/Alpine shell for the AI and the user (Android)
- Kai Build — Coding environment with terminals and coding agents (Android)
- Notifications — Reading other apps' notifications (Android, FOSS build)
- SMS — Reading SMS and drafting replies for the user to send (Android, FOSS build)
- Appearance — Theme picker
- Encryption — Secure settings storage per platform
- Settings Export & Import — JSON backup and restore
- Splinterlands — Auto-battle integration