Siki (式神, Shikigami) is a local agentic AI environment that runs entirely on your machine. Named after the supernatural servants in Onmyōdō tradition, Siki serves as your personal AI assistant with powerful tool-calling capabilities.
- 100% Local - All AI processing happens on your machine
- Single Binary - One portable executable, no dependencies
- Web GUI - Beautiful chat interface with markdown support
- Agent Tools - File operations, command execution, web search, diagram generation
- Twitter AI Curation - Fetch timeline, auto-filter by topic with LLM keyword extraction, evaluate importance (HIGH/MID/LOW), and generate streaming summaries
- Tool Result Pane - Dedicated side panel for full tool output with auto-scroll and click-to-review past results
- Real-time Progress - Live SSE streaming of AI filtering, evaluation, and summarization progress
- Multi-Model Pipeline - LFM2.5 (fast 1.2B) for filtering/evaluation, gpt-oss (20B) for final summarization
- Multi-Backend - Supports Ollama, MLX-LM (Apple Silicon), and vLLM (NVIDIA GPU)
- Image Generation - FLUX.2 Klein 4B for text-to-image generation
- Video Generation - Helios for text-to-video generation
- Mac App - Native .app bundle for macOS
brew tap shi3z/tap
brew install sikiDownload Siki.app.zip from Releases, unzip, and drag to Applications.
# Download binary
curl -LO https://github.com/shi3z/siki/releases/latest/download/siki-linux-amd64.tar.gz
tar -xzf siki-linux-amd64.tar.gz
sudo mv siki-linux-amd64 /usr/local/bin/sikigit clone https://github.com/shi3z/siki.git
cd siki
make build
./siki web-
Install Ollama (if not already installed)
brew install ollama # macOS # or curl -fsSL https://ollama.com/install.sh | sh # Linux
-
Pull a model
ollama pull gpt-oss:20b
-
Start Siki
siki web
-
Open http://localhost:3000 in your browser
siki web # Start web interface
siki web --port 8080 # Custom port
siki web --host 0.0.0.0 # Allow network accesssiki chat # Interactive CLI chatsiki --backend ollama --model gpt-oss:20b web
siki --backend mlx --endpoint http://localhost:8000/v1 webSiki comes with powerful built-in tools:
| Tool | Description |
|---|---|
read_file |
Read file contents |
write_file |
Write/create files |
list_files |
List directory contents |
execute_command |
Run shell commands |
search_files |
Find files by pattern |
grep |
Search file contents |
web_search |
Search the internet |
web_fetch |
Fetch webpage content |
web_images |
Extract images from URLs |
diagram |
Generate diagrams with Graphviz |
twitter_timeline |
Fetch and AI-curate your Twitter timeline |
twitter_search |
Search Twitter with keyword queries |
generate_image |
Generate images with FLUX.2 Klein 4B |
generate_video |
Generate videos with Helios |
run_code |
Execute Python/JS code in sandbox |
| Backend | Platform | GPU Required |
|---|---|---|
| Ollama | macOS, Linux, Windows | No |
| MLX-LM | macOS (Apple Silicon) | No (uses Neural Engine) |
| vLLM | Linux | Yes (NVIDIA) |
# Build for current platform
make build
# Build Mac app bundle
make mac-app
# Build for all platforms
make release
# Create distribution package
make mac-dmgsiki/
├── main.go # Main application
├── web/
│ └── index.html # Web GUI (embedded)
├── macos/
│ └── Siki.app/ # Mac app bundle
├── scripts/
│ ├── build-mac-app.sh
│ └── build-linux.sh
├── Makefile
└── .goreleaser.yaml
Siki can fetch your Twitter timeline, automatically filter tweets by topic, evaluate importance, and generate a curated summary — all locally.
- Configure Twitter API credentials in the Settings panel (OAuth 1.0a)
- Ask: "Twitterのタイムラインからニュースをまとめて"
- Siki will:
- Fetch your timeline via Twitter API v2
- Extract the core topic using LFM2.5
- Generate keywords and filter tweets by relevance
- Evaluate each tweet's importance (HIGH/MID/LOW)
- Generate a streaming summary with gpt-oss
All processing is visible in real-time via the tool result pane.
- Go 1.21+ (for building)
- Ollama, MLX-LM, or vLLM (for AI inference)
- Graphviz (optional, for diagram generation)
- Twitter API credentials (optional, for Twitter features)
MIT License
- Inspired by picoclaw
- Powered by local LLMs via Ollama/MLX-LM/vLLM
- 式神 (Shikigami) concept from Japanese Onmyōdō tradition