Local-first by design
Hosts, sessions, terminal transcripts, credentials, and working context stay on the device whenever the workflow allows it.
Security
Local-first security for real infrastructure work.
VantaShell is built as an SSH workspace where sensitive data, credentials, sessions, and terminal context stay local by default. AI can help, but the user stays in control of context sharing, command execution, and cloud integrations.
Sensitive data
Local-first
Cloud AI
Opt-in
AI commands
Draft only
Credentials
Keychain
Sensitive Mode ready
Local device
hosts, sessions, transcripts, credentials
macOS Keychain
API keys and credentials
Local AI
default assistant mode
Local logs
controlled diagnostics
AI request gate
Cloud off by default
OpenAI or Claude enters the flow only after confirmation and with sanitized context.
Command review
sudo rm -rf /var/log/*
The proposal remains a draft until the user reviews and runs it manually.
Passwords
Tokens
API keys
Private keys
Local AI by default
VantaShell starts with local AI. OpenAI or Claude can be enabled only by the user, with the user's own API key.
Cloud requires confirmation
Any request that sends context to an external provider requires explicit approval and runs through secret redaction first.
Included safeguards
Protection where the work actually happens: terminal, SSH, AI, and credentials.
Security is not presented as magic. These are concrete controls in the app that reduce accidental exposure of sensitive operational data.
API keys in macOS Keychain
AI provider keys and sensitive credentials are stored locally with the operating system's secure storage.
Automatic secret redaction
Passwords, tokens, API keys, and private keys are removed from context before it can be sent to AI.
Sensitive Mode
Sensitive Mode forces local AI and reduces the amount of context that can leave the device.
AI commands stay as drafts
The assistant can suggest commands, but it does not execute them automatically. The user reviews, edits, and confirms each action.
Command security checks
Risky commands are labeled with a risk level and explanation before they reach the terminal.
SSH host key trust flow
SSH host keys are handled through a separate trust flow so important changes are visible before connection.
Separated hosts and sessions
Hosts, sessions, and credentials are managed as distinct entities for clearer control and easier review.
Local logging and diagnostics
Diagnostics stay local for troubleshooting without unnecessary exposure of transcripts or credentials.
Opt-in sanitized AI dataset
Local learning is optional, sanitized for obvious secrets, and does not use raw transcripts as the default source.
An AI flow with visible guardrails.
For simple tasks, VantaShell stays local. For complex work, the user can enable cloud AI, but the data path remains explicit.
01
Local context
VantaShell starts from what is on the device: the session, host, command, and relevant terminal output.
02
Sanitization
Obvious secrets are redacted before any context can be shared with an external provider.
03
Explicit approval
For OpenAI or Claude, the user decides when context is sent and which locally stored API key is used.
04
Reviewable draft
The AI response becomes an editable suggestion or command with visible risk, never automatic execution.
Operational control
The user decides what is saved, exported, and shared with AI.
VantaShell separates sensitive areas so hosts, sessions, credentials, exports, and AI integrations can be understood and controlled individually.
Surface
Posture
Control
Terminal
Controlled transcripts
Exports and sensitive actions remain under user control
SSH
Verified host keys
Dedicated trust flow for host key changes
AI
Local by default
Cloud is optional, confirmed, and sanitized
Credentials
Local storage
Support for macOS Keychain
In short
VantaShell combines SSH/RDP administration with a local-first AI assistant while keeping the user in control.
Data stays local by default, cloud is optional and confirmed, and sensitive commands are analyzed before execution.