Getting started
Install Neuralis, complete first-run setup, and learn the core concepts.
Maturity: stable (90 %)
Host (neuralis). The host provides sign-in, projects with membership, roles and spend limits, package bootstrap and routing, the workspace shell and the operator commands, and leaves every product capability to packages.
- Sign-in is email and password; single sign-on (OIDC, SAML) is not available.
- The host access plane needs a Linux kernel (Linux or WSL2); it is unavailable on macOS and native Windows.
- A package added to the host dependencies runs in-process with full first-party trust; vet it like any dependency.
- Encrypted credentials depend on the master key file; back it up with the data directory.
Maturity: preview (50 %)
Install channels. Neuralis is built to ship through npm create neuralis, a Docker image and a public host repository; none of the three is published today, so installs are built from source.
- No install channel is publicly available; installs are built from the source repository.
- Native installs on Linux compile the terminal module and need python3, make and a C++ compiler.
Neuralis is a self-hosted, multi-LLM agent platform that runs AI agents like
coworkers. From a user's seat it is a browser workspace: you sign in, pick
a project, and work alongside agents that chat, read and write the same files
you do, search a persistent vector memory, run shell commands, drive a real
virtual desktop, and show up on the project calendar with scheduled work of
their own. Everything you see — chat, files, terminal, machine, calendar,
admin — is delivered by packages; the host application only provides identity,
project boundaries, and the workspace shell. That is the platform's one rule,
and it is why anything the AI ecosystem produces — skills, rules, MCP servers,
whole .claude/ folders — can plug into your deployment too.
Who runs it, who uses it
The typical Neuralis deployment is organization-operated: an admin or DevOps team deploys the stack on-prem or in a private cloud, configures model providers and credentials centrally, and invites users. Users sign in through the configured authentication, see only the projects they are members of, and get exactly the capabilities their role grants — every route and tool is checked server-side, deny by default.
That same multi-tenant model scales down to a single self-hoster: one person can be the owner, the admin, and the only user. There is no separate "single-user mode" — the security model is always on.
Two perspectives matter as you read on:
| You are | Start with |
|---|---|
| A user of an existing deployment | Core concepts, then the workspace tour |
| The person deploying Neuralis | Installation, then first-run setup |
How this section flows
Installation
The three distribution channels: npm, Docker image, and git clone.
First-run setup
The interactive setup script: owner account, providers, Qdrant, and your first project.
Core concepts
Projects, agents, packages, contributions, tools, sources, URIs, roles, and sessions.
The workspace
A tour of chat, files, terminal, machine, and the admin dashboard.
Where to go next
Once you are oriented, read the use cases — seven playbooks showing what teams actually build, from a one-person company to a software team. The rest of the documentation splits by audience: the package system is the full contract for building packages, the the first-party packages section documents every first-party package in depth, and the enterprise section covers deployment, configuration, multi-tenancy, roles, and credentials for operators.