Tokengram is a local memory layer for Claude Code. Instead of dumping whole files into the prompt, it finds the exact code your assistant needs — so you burn far fewer tokens and pay far less.
3-day free trial · no credit card · then $3/month
Three steps. No config files, no cloud.
Your repo is embedded into a local vector database on your machine. Nothing is uploaded, ever.
When Claude reaches for a big file, Tokengram intercepts and serves only the relevant chunks via hybrid RAG search.
The VS Code status bar shows tokens — and money — saved in real time, per session and all-time.
The simplest way to cut your AI coding bill — nothing to learn, nothing to change.
A live dashboard in VS Code shows tokens — and real dollars — saved this session and all-time. Most tools hide this. Tokengram puts it front and center.
The index is built and stored 100% locally. No cloud, no uploads, no one reading your repo. Safe for private and company code.
Your assistant gets only the relevant code instead of whole files — often 50%+ fewer tokens on read-heavy work, for less than a coffee a month.
Save a file and Tokengram re-indexes just that file in the background, instantly. Your assistant never works from stale code — and you never click "refresh". Most tools make you re-index by hand, or skip it entirely.
Start free for 3 days — no credit card. Keep it for less than a coffee a month.
No credit card required
Billed monthly · cancel anytime
Works on Windows & Linux (x64) · VS Code 1.85+ · Python 3.12+ · macOS coming soon