NEURALVAULT
CLI
For developers

The fleet, in a terminal.

nv: the same agents, the same memory, the same permission checks Chat uses, run headless. Part of Personal, not a separate purchase.

For The Machines With No One Logged In

Some of this never has a desktop to open.

A scheduled task at 3am. A Windows box you only ever reach over a remote session or SSH. A CI runner that starts, does the job, and shuts down. None of those have anyone sitting in front of them, so an app with a window is not a lighter option there, it is not an option at all. The CLI is not a stripped-down way to use NeuralVault for that machine, it is the only one that fits how it actually runs.

If your own workflow already lives in a terminal, the same is true for you: buy NeuralVault Personal, install it, and never open the app it comes with at all. There is no separate CLI product to choose instead. One licence, one binary, same fleet and memory either way — you are just picking the door you walk through.

Every Surface, From A Terminal

Twelve commands beyond run, not one.

nv run <prompt> reaches the identical engine a typed message in Chat does. The other twelve, transcribed from nv --help, reach everything else.

nv board

The pipeline board: create, move and delete cards. Same AgentFleet gate as the app, same file.

nv orchestra

Put several models on one problem and compare what comes back, cost estimated before it runs.

nv search

Notes matching your words, best first, the same hybrid lexical and vector search the app uses.

nv notes

Browse the vault: folders, ls, read. A direct call, not a model deciding whether to look.

nv sessions

List, read, find, archive, unarchive and rename past conversations from a terminal.

nv audit

The vault’s audit log, newest first: what ran, when, and why it decided to.

nv connectors

Configured MCP connectors, read-only: what is wired up before you script against it.

nv projects

The Projects shelf, read-only, so a script can see what you are already tracking.

nv plugins

List grants and revoke one by id, the same registry the app’s plugin host reads.

nv models

List, browse the catalogue, or set which model a role uses, without opening Settings.

nv keys

Check, set or clear a provider’s API key in your OS credential store.

nv settings

Read local settings: provider defaults, agent-tier overrides, custom endpoints.

Built To Be Scripted, Not Just Typed

A script can tell what happened without parsing prose.

Every tool is denied by default: a headless run has nobody to ask, so anything the model tries that you did not explicitly allow is refused and it carries on knowing that. --allow-tool and --allow-all-tools name what a run may touch, once, in the command that starts it. Exit codes are three states, not two: 0 is a clean answer, 1 is an error, and 2 means the model wanted to do something you had not allowed, which a script can branch on instead of grepping the output for it. --json turns the whole run into one event per line for anything reading it.

Getting It

Not published yet, either on its own or bundled.

A one-line install (irm ... | iex, no download page, no zip to extract) is a familiar shape for anyone who has installed a developer CLI before, and it is the one worth matching. Ours is built and works, and the install command above is real, but nothing is live for it to fetch yet: nv isn’t in any released download today, standalone or inside the app installer.

Gated to Personal at the binary level, not just per command. In developmentBuilt and tested, gated to Personal. Not yet merged to the app’s integration branch, and not in any released installer.

Want the board, the dashboard, chat, and everything else too?