Same kernel, same commands, same output. We run the server, keep your machines in step and give you somewhere to look when you want to know why a memory ranked where it did.
$12 /seat/mo
Monthly. One seat is enough.
$9 /seat/mo
Billed annually, $108/yr. Save 25%.
The free edition already gives one developer everything the kernel can do. Paying starts to make sense at the point where the thing you need is a server that's always up, or a second person.
Self-hosting over MCP gives your laptop a route to one brain. It doesn't give it a copy, so an offline laptop is a laptop with no memory. Hosted replicates: every machine holds the store and reconciles what changed. Two machines editing the same memory is the normal case, not the edge, and the merge is ours to get right.
At two seats a brain can be shared. Memories carry who stored them. Roles decide who can retract a guardrail, and an org-level axiom binds every project the team runs. The line the pricing is drawn on is exactly this: your own data on your own machines stays free, and a store that holds somebody else's work is the paid thing.
Per-brain keys, and the shape that makes the server unable to read what it stores. Skipped in Solo on purpose, because a lone developer has no second party to be protected from and an unnecessary key hierarchy is a good way to lose a brain. The moment a colleague's reasoning is in the store, that argument stops holding.
The local viewer is read-only by design. Hosted gives you the full UI: browse and edit, work the review queue, resolve a conflict the write path parked for a human and see the ranking explained term by term rather than as one number.
An endpoint with OAuth, so you can attach Claude Desktop, the web app or your phone with nothing installed locally. The same 41 tools. No port to open, no certificate to renew, no token pasted into four config files.
| Solo, self-hosted | Hosted | |
|---|---|---|
| Cost | Free | $9-12/seat/mo |
| Projects | unlimited | unlimited |
| Machines | unlimited | unlimited |
| Ranking, decay, supersession, guards | full | full |
| Reaching one brain from several machines | yes, you run it | yes |
| Works with the host down or the laptop offline | no | yes |
| People per brain | 1 | 1 or more |
| Roles, attribution, org axioms | - | at 2+ seats |
| Encryption at rest | - | yes |
| Web UI | read-only, local | full |
| SSO | - | Google, GitHub |
| Standard DPA | n/a | included |
| Who runs the server and the backups | you | us |
The DPA is included at the cheapest paid tier rather than held back for Enterprise. Under GDPR it's table stakes for anyone with EU customers, and a paid tier you can't legally buy from Ireland isn't a tier.
Export is complete and free in both directions. ns-brain export gives you every memory, every event, the supersession links and the provenance as one JSON file, and import reads it back, remapping the ids that don't survive a round trip. Nothing about the paid tier makes your data harder to take somewhere else.
Start free, self-host until the day the constraint is real, and move when it is. That's the intended path, not a fallback.
Signup takes no card. The free edition stays free whatever happens here.