The human interface to CognitiveOS. Two modes: interactive TUI for live sessions, non-interactive CLI for scripting.
- TUI (default): Bubble Tea interactive terminal with 7 display modes, keybindings, voice input
- CLI (
--cmd): Send command, print response, exit.--jsonfor full envelope output.
- Clean text prompt: "Listening..." as default state
- Voice input capture (via ALSA / audio-mcp)
- Text input for keyboard interaction
- Direct framebuffer overlay for images/video (communicates with display-mcp)
- Connects to cognitiveosd via Unix socket
- Non-interactive mode for scripting and automation
make build # compile to build/bin/cognitiveos-cli
make test # run tests
make lint # go vet
make clean # remove build artifactsBoth modes use the same internal/client package to communicate with cognitiveosd via Unix socket. The TUI is thin — it captures input and displays output. All intelligence lives in cognitiveosd and the Wide Model. Either interface can crash and restart without affecting the OS.
cmd/cognitiveos-cli/ Entry point, flag parsing
internal/
├── client/ Unix socket client (shared by both modes)
├── cli/ Non-interactive CLI mode
└── tui/ Interactive TUI mode (Bubble Tea)
github.com/charmbracelet/bubbletea— TUI framework (TUI mode only)github.com/charmbracelet/lipgloss— terminal styling (TUI mode only)- CognitiveOS internal:
cognitiveosddaemon socket
- Use SSH (git@github.com:) for development.
- Use HTTPS (https://github.com/) for build scripts that clone public dependencies.