Open source · MIT · runs on your machine
Give your assistant a memory you own.
imprnt keeps what you know in plain markdown on your disk. You talk, your assistant files what matters, and weeks later it answers from your real history.
The LLM builds the tools. The tools do the work.
# Billing v2
owner: Tom Decker · active
The Q2 priority. No new pricing ships until billing is solid.
- P0: idempotency keys on charge creation, 2-week target
- P0: duplicate-charge metric + pager alert
- P1: invoice redesign (Lena), dunning emails
linked notes
- Its Q2 priority
- Owns it, sets the order
- Invoice redesign (P1)
- Ticket data ranks the backlog
- Lost trust drives the priority
- +3 more linked
The problem
Your assistant forgets everything between chats.
Every session starts blank. You re-explain your projects, your people, and the decisions you already made, and whatever the assistant learned dies when the chat closes.
imprnt keeps the part that lasts in plain files you own. Your assistant reads from your real history every session, and no company can take it away.
How it feels
Talk. It files. It recalls.
Hand over a transcript, a document, or a single fact and ask. The model files structured, linked notes. Weeks later, the answer comes straight from your own vault.
You:Here is my 1:1 with Boris from this morning. [paste the transcript]
imp:Filed it. Created people/boris-carter, updated projects/access-platform with the new cutover date, and logged the meeting under events/.
(three weeks later)
You:What did we decide about the access-platform cutover?
imp:From your notes: the cutover moved to July 15, gated on the two-week parallel-run numbers. Boris owns it. The earlier June date is superseded.
Why it's different
Nothing runs until you run it.
You own the files, you run the commands, and the off switch is real.
Own your files
Plain markdown on your disk, owner-only and local. It opens in any editor, graphs in Obsidian, and traces every note back to its source.
learn moreCheap, local recall
A lookup is about a hundred tokens of local ranked search. No vector database, no server, and it never goes stale on an edit.
learn moreNo daemon, no cloud
No always-on hooks, no auto-injected context, no background loop. The model never takes an action you did not approve.
learn morePlugins you delete
A tiny core plus opt-in plugins you add by name and remove with one line. Compose the assistant you want.
learn moreStart your vault in two commands.
Open source, MIT licensed, and yours to read end to end.