KeepPlain: Memory, Rules and git blame for Your Coding Agents

KeepPlain

6 min read Original article ↗

A plugin for Claude Code, Codex, Cursor, Pi and Antigravity

Keep your coding‑agent sessions. Search them from any agent.

Your agent forgets everything when a session ends. KeepPlain keeps that history as private drafts, so the next session can find what you already solved, follow the corrections you already made, and show which conversation wrote a line.

curl -fsSL https://keepplain.com/install.sh | sh    # macOS and Linux
irm https://keepplain.com/install.ps1 | iex         # Windows, in PowerShell
keepplain login
keepplain enable

Free for personal use. One file, no Node.js, no sudo. keepplain enable asks before it changes anything.

On Windows you can also get it from the Microsoft Store.

Read the plugin’s source (MIT) What leaves your computer

$ keepplain blame src/billing/webhook.ts

● Claude Code · Harden the webhook   4 │ const TOLERANCE_SEC = 300;
○ Pi · Seat count sync              19 │ await syncSeats(event.data, { lock: 'team' });
● Codex · maya, private             22 │ await markPaid(event.data.id, { notify: true });
  no session found                  25 │ return new Response('ok');
git blame, with the agent session behind each line and a link to the turn that made the change. How blame works

You solved this last month. Your agent starts from zero.

With KeepPlain connected, your agent searches the sessions you sent and opens the turn with the answer. Ask in plain words. Git shows what changed, and the session shows why.

How memory works

> how did we fix the webhook retries last time?
> why is public/install.sh like this?
> why was commit 75f9dc0 made?
> /keepplain:resume the seat count sync

Things you can type. The last one picks a task up in another agent.

You corrected it yesterday. Today it does it again.

Once a day a model reads your sessions and suggests the mistakes that came back in two conversations or more. Add one with a click. Your agents get it at the start of a session, or with the prompt it matters for.

How rules work

KeepPlain added 3 rules to this session: Vue (2 yours, 1 from Acme)

Prettier · Vue
Check for the project's Prettier config first.
If there is none, don't run it on shared files.

A rule from our own sessions, and the line your agent shows when rules change.

Claude hit its limit in the middle of the task.

Past 90% of the 5-hour or weekly limit, you get a button to continue in Codex or Pi with a brief: the task, the files changed, the commands that failed, where it stopped. The brief is built on your computer, without a model and without a sign-in.

How handoff works

Claude's 5-hour limit is 92% used, resets 14:30.
Continue in:  [1 Codex]  [2 Pi]  [0 Not now]

The band above the prompt in Claude Code.

What leaves your computer

A tool that reads your agent sessions should say exactly what it does with them.

You decide when

A session is sent only when you run /keepplain:build and confirm, or after you turn on auto mode yourself.

Secrets are replaced first

Keys, tokens, connection strings and email addresses are replaced on your computer. The check works by patterns and can miss things, so you see a preview.

Private until you publish

What arrives is a private draft. Nothing is public unless you publish it.

Where it is stored

On servers in Germany. To pick out the moments that mattered, the redacted text goes to Anthropic’s API. It is not used to train models.

Export or delete

One zip with everything you sent, or delete it all, at any time.

Read the code

The plugin is MIT. Its README lists every network call and what goes with each. The site is not open source. More on security

What it costs

$0

For your own work

Everything on this page, on all five agents. No trial, nothing to upgrade to.

$20

For your team, per active seat a month

Billed yearly, from 3 seats; people who only read don’t count. Rules the team reviews like pull requests, team memory, and the session behind each pull request. Pricing

Questions

Which agents does it work with?

Claude Code, Codex, Cursor, Pi and Antigravity. Some things differ: Cursor and Pi don’t report usage limits, so there you type /keepplain:handoff yourself.

Does a model read my sessions?

Yes, after redaction: to pick out the moments that mattered and to suggest rules. Session text is not used to train models.

Does my agent wait on it?

No. Uploads and rule fetches run in the background. Only Codex, when it exits, waits up to three seconds for the last upload.

How do I remove it?

keepplain disable takes the plugin out of your agents. Delete your account in Settings and your sessions go with it.

Is it open source?

The plugin is, under MIT. The site is not.

Who makes it?

Evgeny Shevtsov. Most of KeepPlain’s code was written in Claude Code sessions. About

Install it, then run keepplain blame on a file your agent wrote.

Questions and problems: [email protected]

More in KeepPlain

For you Free

For your team Per active seat

  • Team rules, reviewedRules your team reviews like a pull request reach every agent on the team.
  • Team memoryAgents search the team’s sessions first, and the work stays when people leave.
  • Sessions on pull requestsEvery pull request shows which commits came from a session, and links it.
  • InsightsWhere agent time goes, what got merged, and at what cost.

For everyone Free