Built by one person, for the long term.

Memrith is built by Caden Bartley. One person, one product, no roadmap pivots. Built because no existing tool would remember the arc of work across years without locking the memory to one AI provider.

Why this exists

Every AI tool I used kept forgetting. Not in a "didn't remember the last message" way — that's solved. In a "didn't remember what I was working on six months ago" way. The chat-by-chat interface doesn't accumulate. The cloud "memory" features are siloed to whichever provider owns them. The "second brain" apps don't talk to AI at all. None of them treat continuity as the thing being built.

The other problem was vendor lock-in. The companies that did try to build long-term AI memory built it on top of one model relationship, one cloud, one set of negotiated terms. The minute you wanted to switch providers, your memory was stranded. The companies were accidentally training people to commit to AI vendors forever — which is exactly the dynamic I didn't want for myself.

Memrith is the version of that tool I would have wanted to exist: local memory that the AI you choose draws on; editable when it gets something wrong; portable when you decide a different model is better this year. The continuity layer belongs to you. The AI provider is replaceable infrastructure.

How it's run

Solo, indie, no investors, no growth metrics besides "are people keeping it for a year." The pricing is one-time so the incentive isn't locking you into a subscription you can't easily cancel. The architecture is local-first so the data isn't a hostage. The roadmap is conservative because the right thing for a continuity layer is to not break the format users have a year of writing in.

The voice across this whole site is intentional. AI products tend to market themselves with words like "revolutionize" and "supercharge" and "10x." I don't write that way and I don't want a product that sounds that way. Memrith is calm because the work it's supporting is calm — long-term thinking, slow accumulation, reflection.

What I'm trying NOT to do

• Add features users don't want, just to look busy.
• Lock users into Memrith the way other tools lock users into them.
• Turn this into a venture-scale company that has to compromise the local-first commitment to satisfy growth.
• Pretend AI is more reliable than it is. Every memory has a provenance trail because AI gets things wrong and you should be able to fix them.
• Spam your inbox. The email list is for shipping notices and the occasional essay, not marketing churn.