Memory for you and your AI

Stop letting AI decide what it remembers.

SiftText is a local knowledge home for you and your agents. It's plain Markdown files you own, with full Git history, and nothing gets saved or brought back without your OK.

Get started See why it's different

A binder, not post-it notes.

Every AI chat is a sheet of paper. There's only so much room, and it gets thrown out when the chat ends. What matters is what you keep and who chooses it.

📄

The paper

The model's context window. It has limited space and resets every chat. Once it fills up, the model loses the thread.

🟨

The post-it notes

Automatic memory. The system picks what to write and when to show it. The notes take up space on the page, and you can't see what it chose or why.

📒

The binder: SiftText

You save the whole page or just the part that matters, and you name the sections. The model suggests what to file, and you approve it.

Why people switch to SiftText

It's built on tools developers already trust, and it puts you in charge.

01

You approve every save

Your agent shows you the exact change it wants to make and waits for your yes. Your memory never changes behind your back.

02

Plain Markdown on your machine

Your knowledge lives in readable files in a folder you control, not in some vendor's hidden database. Open it in any editor.

03

Full Git history

Every approved change is a checked Git commit, so you can see what changed, when, and why, and you can undo it.

04

Works with your agents

Point Claude, Codex or Pi at your knowledge home with a one-line entrance. Your memory goes with you from one tool to the next.

05

Search that finds things

Built-in search covers your whole home, so your agent opens the right page instead of guessing or loading everything.

06

Checks before anything is published

Edits happen in a private worktree and pass checks before they're submitted. A messy draft never ends up in your binder.

Brain, Tree, Beam

Your brain is great at seeing connections but can't write them all down neatly. A language model can go deep on one focused question. SiftText's tree of small, linked pages connects the two.

  1. Get it out of your headAn agent interviews you and pulls the pieces out of your messy thinking.
  2. One idea per pageEach piece becomes a page with a clear scope, small enough for you and the model to think about clearly.
  3. The model goes deepIt puts all of its attention on that one page.
  4. You check itYou confirm the page says what you actually think.
  5. The tree remembersThe page sticks around and gets richer over time. The model doesn't have to remember, because the tree does.

"The language model does not need to remember. The tree remembers."

Get started in a few minutes

Install SiftText, create your knowledge home, and point your agent at it.

# sign in and activate
sifttext login

# create your knowledge home
sifttext init ~/sifttext

# find what you already know
sifttext search ~/sifttext "onboarding"

# make edits in a private worktree, then submit with checks
sifttext worktree ~/sifttext