Terminals
Native shells, agent terminals, history policy, and supervision.
Terminals in BLXAde are native PTY sessions managed by the desktop runtime, not simulated browser consoles alone.
Shell terminals
Discovery
At startup BLXAde detects platform shells:
| Platform | Catalog |
|---|---|
| Windows | Windows PowerShell, PowerShell 7, Git Bash, Command Prompt |
| Linux / macOS | Bash, fish, system shell, Zsh |
Only detected, available shells can be enabled for the canvas new node menu. Unavailable shells stay visible in Settings but cannot be toggled on.
Settings → Shells
- See the detected OS and each shell’s availability
- Enable or disable which shells appear under Add → Apps → Shell
- Rescan if you installed a shell after launch
Disabling a shell removes it from the menu only; existing nodes can still run.
Theme injection
New shells receive theme-related environment values so embedded TUIs can match BLXAde appearance. Light mode can optionally dim terminal palettes while the rest of the UI stays light (Settings → Appearance).
Agent terminals
Agent CLIs discovered under Settings → Agents launch as agent terminals. Some providers (e.g. Codex, Kimi, Antigravity) attempt session binding for resume for up to ~120s beside the PTY. That observer is best-effort: a timeout leaves a healthy process running unbound. Only a concrete persistence failure after a candidate is found may error the node. PTY exit remains the authoritative lifecycle signal.
Header actions for agent terminals include Handoff to the Global Agent (Agents).
Context compaction
When idle and supported, agent terminals can run provider-specific
/compact / /compress style compaction. Antigravity may expose this as
disabled. New-session reset feedback uses the normal session lifecycle.
History and privacy
| Layer | Behavior |
|---|---|
| Full history | Held in process RAM on the native side |
| UI replay | Bounded buffer for the embedded xterm view |
| Agent reads | ANSI normalized; secrets/tokens/private keys masked |
| Persistence | Full transcript is not written into the canvas file |
When the app exits, RAM-only history is gone unless you copied or handed off content yourself.
Reactive supervision
Agents and tools can wait for terminal output via reactive wait/follow tools instead of busy-polling. Multiple agent terminals can be supervised in parallel.
Practical tips
- Prefer agent terminals for CLI coding agents; use shell terminals for plain builds, servers, and scripts.
- Keep secrets out of commands when possible — masking helps but is not a guarantee against all leak shapes.
- Use Screenshot on the node when you need a visual share without raw logs.
- If a shell is missing from the menu, open Settings → Shells → Rescan.