Open Settings from the Manage menu, as in the Editor. The Agents app reads the same Neural Inverse settings and managed policies as the Editor; this page lists the settings specific to sessions and engines.
Sessions and worktrees
| Setting | Default | Meaning |
|---|---|---|
sessions.worktreeLocation | besideProject | Where new session worktrees are created: besideProject (a <project>.worktrees folder next to the project), insideProject (a .worktrees folder inside the project, excluded from git) or custom |
sessions.worktreeCustomFolder | empty | Absolute folder for worktrees when the location is custom; each project gets its own subfolder. A custom choice without an absolute folder falls back to besideProject. |
sessions.autoArchiveOnPullRequestClose | false | Archive a session once every pull request it opened is merged or closed |
sessions.showApplicationBadge | off in stable builds, on in pre-release builds | Show a badge with the number of unarchived sessions on the dock (macOS), launcher (Linux) or taskbar (Windows) icon |
sessions.applicationBadge | Which unarchived sessions the badge counts (for example those that need input, have failing CI, or are unread) | |
sessions.chat.compareAgents.enabled | false | Experimental. Show Compare Models in the new-session model picker |
Engines and models
| Setting | Default in Neural Inverse | Meaning |
|---|---|---|
chat.agentHost.claudeAgent.enabled | true | Register the Claude engine. Needs an agent host restart. |
chat.agentHost.codexAgent.enabled | false | Register the Codex engine. It runs only on a ChatGPT or OpenAI Codex account (sign in under Subscriptions on the Models page), never on connected models. |
chat.agentHost.byokModels.enabled | true | Offer your connected models to new agent sessions |
Automations (experimental)
| Setting | Default | Meaning |
|---|---|---|
chat.automations.enabled | true | Show Automations and dispatch scheduled runs. When off, scheduled runs are not dispatched; definitions are kept. |
chat.automations.runTimeoutMinutes | 30 | Maximum minutes a run may take before it is ended. Offered only in pre-release builds. |
Environment variables
| Variable | Meaning |
|---|---|
NI_SHARED_AGENT_HOST | 1 forces one agent host shared by both apps; 0 gives each app its own. Default: shared on macOS, separate on Windows and Linux. |
Appearance
The Agents app always uses its own modern design. workbench.experimental.modernUIEditorTabStyle (connected or pill) applies to its chat and side-panel tabs, and workbench.modernUIFrostedGlass and workbench.modernUIFrostedGlassOpacity to its menus, pickers and dialogs. See Settings and policies.
Shared with the Editor
The Agents app reads the same sign-in, governance, audit, context, vault and hardening settings and managed policies as the Editor, including NeuralInverseKeystoneBaseUrl for the Govern deployment, and applies the same rules about who may set what (several are managed policy only). On macOS its configuration profile targets the bundle identifier com.neuralinverse.agents. The full list is in Settings and policies.
Under an organisation policy, the agent auto-approval settings (chat.tools.global.autoApprove, chat.tools.edits.autoApprove, chat.tools.urls.autoApprove, chat.tools.terminal.enableAutoApprove, chat.tools.terminal.autoApprove) are governed, so a value in a settings file is ignored.