Default route
Local Only
Terminal output, failed commands, summaries, search, and quick explanations stay on the device by default.
AI Safe Local
Local-first AI for safer SSH decisions.
VantaShell includes a terminal assistant that explains output, diagnoses failed commands, drafts next checks, and reviews command risk without turning AI into an automatic executor.
It runs locally by default. OpenAI and Claude are available only when the user enables them with a personal API key, approves the handoff, and lets VantaShell redact sensitive context first.
Default
Local Only
Cloud
Explicit opt-in
Secrets
Auto-redacted
Execution
Never automatic
Routing
Three clear AI routes, one visible boundary.
The assistant can stay fully local, work in hybrid mode, or use a cloud provider. The important part is that the route is explicit before sensitive terminal context leaves the device.
Local first, cloud by approval
Hybrid
VantaShell prepares the context locally, redacts sensitive values, then asks before a selected task is sent out.
User-owned provider keys
Cloud Only
OpenAI or Claude can be enabled for complex work with the user's own API key and a visible provider boundary.
Assistant capabilities
Built for understanding, diagnosis, and safer preparation.
AI Safe Local focuses on helping the operator understand what happened and prepare the next move. It does not silently run commands, change servers, or hide provider decisions.
Explain output and logs
Turn command output, stack traces, service errors, and log fragments into short operational explanations.
Suggest failed-command fixes
Read the last failure, identify likely causes, and propose the next checks without running them automatically.
Draft diagnostic commands
Generate reviewable commands for disk, RAM, processes, logs, nginx, containers, Kubernetes, networking, and permissions.
Check command risk
Flag destructive, broad, privileged, or production-sensitive commands before the user decides what to do.
Summarize sessions
Compress long terminal output or a whole session into decisions, warnings, and next actions.
Search the transcript
Find the command, output block, error, token-shaped value, or service name that matters in the active terminal history.
Copy and controlled rerun
Copy the last command or last output quickly, then rerun only after VantaShell has surfaced the command risk.
Promote useful patterns
Suggest snippets and macros from repeated commands so everyday SSH work becomes faster and easier to review.
Diagnostic drafts cover the daily areas where SSH work gets noisy: resources, services, containers, orchestration, networking, and permissions.
DiskRAMProcessesLogsNginxDockerPodmanKubernetesNetworkPermissions
Safety model
Risk review before action.
The assistant is a copilot for terminal work, not an unattended shell. Suggestions are kept reviewable, destructive commands are flagged, and cloud handoffs are explicit.
VantaShell can help prepare commands such as reruns, diagnostics, snippets, and macros, but the user stays responsible for reviewing the final command before it touches a live host.
01
Analyze locally
The assistant starts from terminal context already available on the Mac.
02
Redact secrets
Passwords, tokens, API keys, private keys, and similar sensitive values are masked before sharing.
03
Review risk
Commands that can delete data, change permissions, stop services, or touch production get an explicit warning.
04
User approves
Suggestions remain drafts until the user reviews, edits, and chooses the next action.
Privacy controls
External AI is optional, deliberate, and masked.
When a more capable model is useful, VantaShell keeps the handoff visible: choose the provider, use your key, confirm the context, and send only after redaction.
No invisible cloud path
External providers are disabled until the user turns them on for the workflow.
Provider consent prompt
When OpenAI or Claude is needed, VantaShell asks before sending terminal context outside the device.
User API keys
Cloud assistant access uses the user's own provider credentials, not a hidden shared key.
Opt-in local dataset
Local data collection for future AI adapters is deliberate, visible, and separate from daily assistance.
Bottom line
A terminal copilot that helps you think before you run.
AI Safe Local is designed to explain, diagnose, summarize, draft, and warn while keeping execution under human control and terminal context local unless the user approves otherwise.