Lean · Pro · for AI agents

Stop paying for code your agent already read.

Your AI agent re-reads the same files all session. Unchanged repeats are served from memory instead of being sent and billed again. Works with Claude Code, Codex, Gemini CLI and GitHub Copilot CLI.

Start free See pricing

Runs locally. Nothing changes about how you work.

claude code · prompt genie active

natik@mbp shop-api % npx prompt-genie [prompt-genie] lean on > session refresh fails after ~15 min, find out why ▶ Read src/server/auth.ts 420 lines, ~3,800 tokens sent to context ▶ Read src/server/session.ts ▶ Edit src/server/session.ts (+7 -3) ▶ Bash npm test -- session 14 passed, 0 failed ▶ Read src/server/auth.ts (again, 5 turns later) ✓ Already in context, unchanged since your last read. 420 lines.

Tokens skipped this session 129,280shop-api (main)

Lean Pro

Tokens skipped this session

129,280

About two thirds of a context window, never sent again

src/db/schema.ts
Re-read 8 times · unchanged, from memory 64,920

src/server/auth.ts
Re-read 11 times · unchanged, from memory 41,800

npm test -- session
Same output 6 times · from memory 22,560

src/server/session.ts
Edited since last read · sent fresh 0

What you pay for without it

Your agent re-buys context it already has, all day.

3,800+: tokens re-billed every time your agent re-reads one file
10x+: the same file gets re-read across a long session
70%: of your repeat-read tokens are avoidable waste

How it works

Memory for your AI agent, in three steps.

Install once and it runs quietly on every session.

01
Install in one command
Run npx prompt-genie, enter your email, restart your agent. It finds Claude Code, Codex, Gemini CLI and GitHub Copilot CLI on your machine. No config files, no API keys.
02
It remembers every read
Hooks fingerprint each file and read-only command your agent reads this session, so it recognizes the exact moment something is read again.
03
Repeat reads cost almost nothing
Unchanged content that is already in context gets served from memory, not sent to the model again. You stop paying for what your agent already knows.

Runs locally, your code never leaves your machine
Per session and compaction aware

Questions

Questions, answered.

Anything else? Contact us and a person will reply.

No. The check runs after the read, comparing a hash of the fresh output to what was already delivered this session. If a file changed in any way, it passes straight through untouched. There are no caches to go stale.