Settings map¶

Where each setting lives. The settings overlay has seven pages, listed down the left sidebar in this order.

Page	Subtitle in the app	Groups on the page
General	App behaviour, fonts and layout	Appearance, Layout, Behaviour
Terminal	Shell appearance, scrollback and cursor	Display, Interaction, Advanced, Local Terminal
AI Providers	Connect Claude, ChatGPT, Gemini, Groq, Ollama and more	Provider cards, Provider Behavior
AI Assistants	Connect local assistants and ordinary ChatGPT chats to OpsPilot's approval-gated MCP tools	Use an AI assistant you already have, Assistant connections, Session access, Advanced
Environments	Add, rename, recolor or remove the environments connections can be tagged with	Environments, Add environment
Security	Idle timeout and session lock policy	Idle Lock, Credentials, Command Safety, Data Handling
About	Product, company and licensing information	Product hero, NubeStack attribution, Security guarantees

Each page saves with its own button — Save General Settings, Save Security Settings and so on. Nothing is saved until you press it.

General¶

Group	Control	Default
Appearance	Theme	OpsPilot Dark — the only entry
Appearance	Font family	JetBrains Mono, or Menlo on macOS
Layout	Left panel default width	260 px (160–600)
Layout	Show status bar	On
Layout	Show breadcrumb bar	On
Behaviour	Confirm before closing a session tab	On
Behaviour	Reopen last sessions on start	Off

One theme ships and there is no theme selector, so Theme has a single option.

Terminal¶

Group	Control	Default
Display	Font family	As General
Display	Font size	13 (8–32)
Display	Line height	1 (1.0–3.0)
Display	Scrollback buffer	10000 lines (500–100000)
Display	Cursor style	Bar (also Block, Underline)
Display	Cursor blink	On
Interaction	Copy on select	On
Interaction	Right-click pastes clipboard	On
Interaction	Bell	None (silent), Visual flash or Sound — no option is pre-selected
Advanced	Minimum terminal height	220 px (100–800)
Local Terminal	Default shell	Filled in per platform

Scrollback is held in memory only. OpsPilot never writes it to disk.

AI Providers¶

One card per provider, ten in all, each with its own credential fields, a Test button and a model picker. Below the cards is a Provider Behavior group whose settings apply only when you are using a pay-as-you-go provider — an assistant connector is pull-based and is not affected by them.

Auto-run safe commands is not on the AI Providers page. It sits under Settings → Security → Command Safety, because it is a command-approval policy that applies to every command source rather than a provider-specific behaviour. The AI Providers page says so in place, and links across.

Full detail: AI provider matrix and AI providers.

AI Assistants¶

The MCP connector and the external apps that use it.

Group	Control	Notes
Use an AI assistant you already have	Enable connector	Off by default. Starts the local listener
Assistant connections	Claude Desktop, VS Code (Copilot Chat), ChatGPT	Each has Connect / Disconnect and a status badge
Assistant connections	Tunnel	Configure opens a panel for the tunnel ID and runtime API key
Session access	Allow opening/reconnecting sessions	Off by default
Advanced	Server URL, Access token	Read-only, for manual or debug connections

Allow opening/reconnecting sessions is the master switch for open_session. It only applies to connections that also have Let AI Assistant open this session turned on, which is off by default per connection. See MCP tools and AI Assistants.

Environments¶

Add, rename, recolour and remove the environment tags a connection can carry. An environment is a colour-coded tag, not a workspace. See Organising connections.

Security¶

The page that a security reviewer reads. Four groups.

Idle Lock¶

Control	Default
Lock after idle	10 minutes (1–120)
Lock on window minimize	Off

Credentials¶

Control	Value
Credential storage	OS Keychain, shown as a badge, not a choice
Clear all saved credentials	Clear button. Cannot be undone

Where each kind of secret is held:

Secret	Where it is stored
SSH passwords, private key passphrases	`credentials.enc.json`, encrypted by the operating system
OpenStack passwords and application-credential secrets	The same encrypted path
S3 secret access keys	The same encrypted path
OpenAI tunnel runtime key	Its own encrypted file, decrypted only into the tunnel process
AI provider API keys	`settings.json`, in plain text

Connection credentials are fail-closed: if the operating system cannot encrypt the credential, saving fails rather than storing it in the clear. Provider API keys take the ordinary settings path instead — see Security model for what that means for a review.

In 0.1.0, Clear removes saved AI provider configuration, including provider API keys. It does not remove saved connection credentials; delete the connections that hold them.

See Credentials.

Command Safety¶

The rows here edit the Default Command Safety profile — the one used by any group or connection with nothing more specific assigned.

Control	Default	What it does
Auto-run safe commands	Off	Read-only commands the model marks safe run with no approval click
Dangerous command confirmation (Default profile)	Toggle	On: a dangerous command needs a typed justification of at least ten characters. Off: it still needs an explicit click
Command Safety profiles	Manage profiles…	Edit the Default list, clone new profiles, assign per group or per connection
AI direct execution	Always enforced	A badge, not a toggle

The approval buttons on a proposal card are approve & run and dismiss.

Auto-run safe commands applies to proposals from a configured provider and from a connected AI Assistant alike — one policy for every command source. It is a single application-wide switch: it is not per profile, per group or per environment. Per-environment strictness is what Command Safety profiles are for. The page states the reason it ships off: it trusts the model's own judgement on each command, and the dangerous pattern list is the only thing that independently overrides it.

AI direct execution is reported as Always enforced. The model never touches a real shell; OpsPilot's own code is what runs anything, and that boundary cannot be changed. Turning Auto-run safe commands off or on changes whether a click is needed, never who executes. Dangerous commands always require an explicit click.

See Command Safety profiles, Dangerous patterns and Execution boundary and risk tiers.

Data Handling¶

Control	Default	What it does
Redaction	Always on	Secrets are scrubbed from terminal output before it leaves the app
Show redaction badge	Toggle	Purely a display choice
Data Handling profiles	Manage profiles…	Category toggles and custom patterns per profile

Hiding the badge does not disable redaction. See Data Handling profiles and Redaction categories.

About¶

Product name and version, the NubeStack attribution and a link to the company site, a Security guarantees list, and the copyright footer.

One line in the Security guarantees list is loosely worded: it says dangerous commands require "a typed confirmation phrase". The real gate is a free-text justification of at least ten characters, which doubles as a record of why the action was approved. There is no phrase to type.

Identifying the packaged version

The released version of this build is 0.1.0. Identify what you are running from the download or installer you ran.