The default is someone else's cloud
As AI agents take on real work, they generate a new kind of data: plans, decisions, memory, the reasoning behind a change. The convenient default is to ship that to a hosted service. It is easy to start, easy to scale, and it quietly means your most sensitive context — your codebase, your project's history, your working notes — lives on infrastructure you do not control.
For some teams that trade is fine. For agent infrastructure, it usually is not. Agent memory and collaboration are not a feature you rent; they are the layer that understands your work. Owning it matters.
What "local-first" actually means
Local-first is not the same as offline or un-networked. It is a set of commitments about where your data lives and how you can use it.
Your data stays with you
Code, sessions and memory sit on your machine, in a format you can read, back up and move. No account, no third-party storage, no silent uploads.
You can see the machinery
Plain files, a SQLite database, a local daemon. When something looks wrong you can open it, not file a support ticket and wait.
It fits what you run
Open protocols like MCP, standard files on disk, no lock-in. The tools work with the agents and workflows you already have.
The trade-offs, honestly
Local-first is a choice, not a free lunch. Running software yourself means you install it, you update it, and you back it up. You do not get a managed service that scales for you or stays up because someone else is on call at 3 a.m. A hosted product will often be faster to try and easier to forget about.
What you get in return is exit. Your memory is a file, not a hostage. Your collaboration state is a directory you can delete without asking permission. For infrastructure that sits under your work, that is usually the right trade.
How Alionix builds it
Both Alionix products follow the same principle in different jobs.
neoHive — collaboration without a server
neoHive coordinates AI agents with no central service at all. Each tool spawns its own MCP process, and they all share a project-local .neohive/ directory of append-only files. The filesystem is the message bus. Read the neoHive overview →
neoBrain — memory that never leaves the host
neoBrain keeps an agent's memory in a single SQLite database and runs embeddings on-host by default, so memory text does not leave the machine. Its daemon binds to 127.0.0.1, and every retention decision is deterministic and reversible. Read the neoBrain overview →
Frequently asked
Does local-first mean I cannot use a cloud model?
No. It describes where your data lives and how the tool runs, not which model answers. You can point the tools at any model endpoint you like — or use none, for storage and retrieval.
Is local-first the same as self-hosted?
Related, but stricter. Self-hosted means you run the server. Local-first means the data model is designed for your machine first — single files, local daemons, no vendor account required even to start.
Can I still back up or sync my memory?
Yes — that is the point. The store is a file, so ordinary backup and sync tools work on it without any special integration.