NeuralVault gives you the pieces: Orchestra to split one job across several models, your own MCP servers, plugins and skills, and an approval card on every write.
An illustration of a screen, with made-up content, not a screen capture.
Every write to your vault waits on an approval card showing which files it touches, so a chain of agents cannot change your files unseen.
Orchestra splits a task across models and combines the answers, with an estimate before it runs.
Add the MCP servers, plugins and skills you already use, and switch them on or off from Settings.
Use your own key or a local model for each part of the job.
Orchestra puts them on parts of one question and combines the answers.
Anything written to your vault waits for a yes.
Writes wait for a yes, and one control puts the files back after an approved chat write.
Put several models on one problem. Each agent runs in its own lane, so you can see which did what.
Add a connector, give it its credentials, and press Check: it starts the connector and reports what answered.
Install a plugin, switch it off or remove it, and teach agents skills that live as files you can read.
Point NeuralVault at a DGX Spark, a home server, vLLM, LM Studio or a rented box, several at once.
Checks each action before it runs, so a step that goes wrong is stopped or asked about before it does damage.
See what it catches, with the scoresWorkflows are not in the app yet, and an approvals queue for work that runs while you are away is not part of 2.0. The nv command line, for scripts and cron, is coming soon and is not in a public download yet. Free is chat first: the Dashboard, Library, Orchestra, MCP connectors, plugins and the agent features need a Personal plan.
One email when a NeuralVault version goes out. We note which page you signed up from, so we know who to write for.