For developers

Let agents do real work in your repo,
on your rules.

If you want an agent you can supervise, this one is built for you. NeuralVault runs on your machine: every command and change waits for your approval, a guard with fixed rules checks each action, and every run is recorded in a log you can read.

See editions and pricingDownload NeuralVaultGet release updates by email
Not your role? See NeuralVault forDevelopersAI and automation buildersIT and securityDirectors and executivesFounders and solo operatorsManagers and ownersOperationsFinance and procurementMarketing teamsResearchers and analysts

An illustration of a screen, with made-up content, not a screen capture.

Why it fits you

You want agents that do real work in your repo, on your terms, without surprises.

  1. 01

    You stay in control

    Pick how much it may do on its own: Plan, Manual or Default. Every command and every change waits for your approval, and deletions are always refused.

  2. 02

    Your standing rules go in first

    Write standing rules into CORTEX.md. NeuralVault gives them to the agent at the start of every run. Separately, a guard with fixed rules checks every action.

  3. 03

    Nothing to be locked into

    Notes and memory are plain Markdown on your own disk, and the audit log is a plain text file, so your own tools can read them.

The difference it makes

Before, and with NeuralVault.

Without it

  • An agent that acts first and explains later
  • Instructions that live only in your head or in a chat
  • No record of what actually ran

With NeuralVault

  • Every command and change waits for your approval
  • Standing instructions in a file, read in at the start of every run
  • Every run logged, so you can read it back
  1. 1

    Write the rules

    Put them in CORTEX.md. They are given to the agent at the start of every run.

  2. 2

    Approve as it goes

    Commands and changes wait on a card that shows what they will do.

  3. 3

    Read the log

    See what ran, and what the guard stopped.

What it does for you

Built by people who wanted
to read what the agent did.

Approvals you control, runs you can read

Choose Plan, Manual or Default, and commands and changes wait on a card that shows what they will do. Everything it does goes into an audit log.

Standing instructions, in a file you own

Settings shows which file the assistant is obeying (CORTEX.md, CLAUDE.md or AGENTS.md), exactly as it is on disk.

Code mode that reads your instruction files

When a turn used your project’s own instruction files, the page says so, and says it more strongly when that text read like orders.

Bring your history with you

Import your conversations from Claude Code or Codex, so you start with context, not a blank page.

Your model, your key

Point it at your own key or a local model and nothing is metered. Any MCP server you already run still works.

A guard on every action

It checks each action before it runs for leaked keys, look-alike installs, downloads piped into a shell, known code flaws and instructions hidden in something it read.

See what it catches, with the scores
What is not ready

NeuralVault is Windows only for now. The nv command line, for scripts, CI and cron, is coming soon: it is built and tested and not in a public download yet. Sync across your own devices is in development, and macOS and Linux builds are not part of 2.0. Free is chat first: the Dashboard, Library, Orchestra, MCP connectors, plugins and the agent features need a Personal plan.

See the roadmapEditions and pricing
Release updates

Get an email when NeuralVault ships for developers

One email when a NeuralVault version goes out. We note which page you signed up from, so we know who to write for.

Almost there. Check your inbox and click the link to confirm, and you are on.

Double opt-in, and one click to stop. Only the NeuralVault release list.