One persona, any platform

Nothing forgets.
Nothing forks.

Keep your custom persona and everything it has learned in one place you own — then carry it across any app, any model, hosted or local. The memory lives outside the model, so switching the model never costs you the history.

portable personamodel-agnosticread · write · summarizeone source of truth

Your persona is portable

01You bring your agent with you

The persona you build and the past it carries aren't trapped inside one product. Point a fresh app or a different model at the same memory and it picks up mid-sentence — same voice, same context, same history. You bring your agent with you instead of starting over.

The coin below is the persona. Swap the platform frame around it — hosted, local, any app — and watch what stays put.

hosted workshopcloud model
local machineyour own model
any appany model
your persona · unchanged

It lives outside the model

02The model is swappable. The memory isn't.

Memory isn't baked into whichever model happens to be answering. It sits in one canonical place that every model reads from and writes to — so the model is swappable, upgradeable, replaceable, and none of that touches what's remembered.

why it matters

When memory lives inside a model, every upgrade is a small amnesia. Move it outside, and the model becomes a part you can change without loss.

model Aswappable model Bupgradeable model Creplaceable the memory one canonical place

Read, write, summarize — model-agnostic

03One persona and canon, feeding many frames

Three plain operations: recall what happened, record what's new, distill the long story into the short one. Any tool that can reach the memory can use them — and they work the same no matter which model is on the other end.

hosted workshop cloud model reading & writing your own machine local model, offline same history any app third-party tool picks up mid-sentence a fresh model brand new, day one already knows you read write read + write summarize one persona · one canon
One persona and its canon in the center; every platform frame reads, writes, and summarizes against the same store. The container swaps — the identity holds.

read

Recall what happened. Any tool asks the memory what it already knows and gets the settled record back — not a guess reconstructed on the fly.

write

Record what's new. A fact learned in one app is written once to the shared store, and every other tool sees it the next time it reads.

summarize

Distill the long story into the short one. The years-scale record compresses to what matters now, so recall stays fast as the history grows.


One source of truth

04Can't be talked into forgetting or lying

Because the whole stack reads and writes the same store, there's a single settled record underneath everything. Two tools can't quietly drift into disagreeing about what happened — they're both looking at the same canon.

What's true is what's written down and provable, not what a model feels like saying this turn. The memory is the authority: it can't be coaxed into dropping a fact or inventing one, because the record decides — not the phrasing of the question.

the tie that holds the stack together

For building and for keeping things running, this is the connective layer — the place every other tool checks against, so nothing goes stale and nothing loses the thread.

tool onebuilding tool tworunning one settled record no fork · provable canon