SiftText
memory you govern
Jack ten Bosch
Pitchtank SF
$20
millions of tokens
lost
two weeks ago i was too lazy to use my own product. i ran deepseek on a small project and eighty percent of the budget went to going in circles. now it sits unfinished, and i can't explain it to a fresh chat, because i can't explain the same thing twice the same way.
the chat

a piece of paper

dumb and limited. you fill it, then throw it out when the next chat starts.

auto-memory

post-it notes

the labs stick notes on the paper for you. they pick which ones, they cover space you needed, and you don't see what they chose.

siftext

a binder

save the whole page or the part that matters. you name the sections. the model files and fetches under your gate.

a graph database where the llm is its own hopper. it does its own context engineering.
the right, never the obligation
pushforce context into the window when you want it
pullgrab the exact node on command
readsee what you stored, full observability
writeorganize by hand, or let the model do it
auto-memory forces the push and hides the read. obsidian forces the write. siftext hands you all four and runs them under your gate when you trust it.
doom on android
175 nodes filed. the model opens the slice it needs, not the binder.
~9,000nodes across real projects
5paying users
$100mrr
$250one-project memory rebuild zoom
i run big projects and watch what the model does the whole way. that is peace of mind. the trees i build today are the training data for the model that runs them tomorrow.
find me after.
sifttext.com
esc to leave demo