Keep one context. Every session, every AI, every teammate.
Every AI session starts blind. You re-explain the same things — what the project is for, how
to log in to your servers, where the logs live, what was decided last week and why. KeepCTX
gives your project one file where your AI writes those things down as it learns them, and
reads them back at the start of every session.
curl -fsSL https://keepctx.com/install.sh | sh
cd ~/work/your-project && keepctx init
# optional: commit KEEPCTX.md so teammates and their AI agents see and update the context too
git add KEEPCTX.md AGENTS.md && git commit -m "Add KeepCTX"
That's all. Open source, no account, no server.
What keepctx init does
- Creates
KEEPCTX.md in your project — a plain Markdown file of facts, with the rules for
keeping it at the top.
- Adds a short pointer to the top of
AGENTS.md, the file Claude Code, Codex and most
other agents already read: read KEEPCTX.md before your next reply of any kind and
keep it current; if it's missing, ignore this and carry on. Nothing else in AGENTS.md is
touched.
From then on your AI reads and edits the file like any other. Nobody needs KeepCTX
installed — not your teammates, not their agents. The rules travel inside the file.
Stats
Installer downloads per day, for the last 30 days and the last year:
keepctx.com/stats.html. Counted from the website's access logs —
keepctx itself never sends anything anywhere.
Updating
Re-run the install command, then keepctx init in each project:
curl -fsSL https://keepctx.com/install.sh | sh
cd ~/work/your-project && keepctx init
keepctx init replaces the rules — everything above ## Facts — with the current version and
never touches the facts. Commit the result and your teammates get the new rules through git. Keep
your own instructions in Facts or AGENTS.md, not in the rules section, since an update replaces it.
What it looks like
## Facts
- **logs.location** — Splunk, index app_prod. Not CloudWatch.
- **prod.access** — SSO, then the bastion host. Never direct SSH.
- **storage** — SQLite: one file to back up, no database server to run
- **rejected.postgres** — a server to operate for a few megabytes of text
Whenever the AI changes the file, it tells you in one line:
KeepCTX: added — storage: SQLite (or "updated", "removed").
What goes in it
Whatever the AI judges worth keeping. A wide net: not just how things run, but what's being built and why. Never secrets, never
what you put off the record. It isn't only for code — anything you keep re-explaining to an
AI belongs in a context.
Sharing
It's a file in your repo, so it's shared the way your code is: commit it, and teammates get it
when they pull. History, review and revert work as they do for everything else.
Contributing
src/keepctx.py is the whole tool: one file, standard library only. The template and the
rules in it are the product, so changes to them matter most. See
CONTRIBUTING.md, and DESIGN.md for why it's this simple.
License
Apache-2.0.
