Tokengram — Token-saving memory layer for AI coding assistants

Stop paying for AI to re-read your codebase. Tokengram indexes your project locally and serves only the relevant chunks to Claude Code (and other MCP-capable assistants) via hybrid RAG search.

Measured token savings: ~70% on a typical Python project. Local-only. Your code never leaves the machine.

Free 7-day trial, then $2/month to keep the license active. Subscribe at tokengram.click. One key per machine; cancel anytime.

Why Tokengram

When you ask Claude a question, it normally reads whole files into its context window — 5–10K input tokens per query. Tokengram pre-vectorizes your project and returns only the 3–5 most relevant chunks when Claude asks. So:

⚡ ~70% fewer input tokens → cheaper API calls
🔒 Local-only — embeddings + search run on your machine
♻️ Auto re-index — saved files are picked up automatically
🎯 Hybrid search — semantic (meaning) + BM25 (keyword) + cross-encoder reranker
📊 Live status bar — see real-time savings as you code

When it works

✅ Medium-to-large Python / JS / TS / Go / Rust / C# projects (50+ files)
✅ Contextual questions: "where is this function called from?"
✅ Refactoring (gathering all touch points)

When it doesn't (yet)

❌ Projects with < 20 files — Claude can just read them all
❌ Machines with < 4 GB RAM
❌ Systems without Python 3.12+ installed

Quick start

Install this extension from the Marketplace.
Open a workspace and run Tokengram: Open menu from the Command Palette (Ctrl+Shift+P).
Follow the setup wizard:
- Tokengram detects Python 3.12+
- Creates an isolated environment in ~/.tokengram/
- Asks for your license key
- Configures Claude Code hooks for the workspace
Open Claude Code and start asking questions — savings appear in the status bar.

Requirements

VS Code 1.85+
Python 3.12 or newer (download)
Claude Code (or any MCP-compatible assistant)
License key — $2/month, get one at tokengram.click

Status bar

A live odometer in the bottom-right shows tokens saved in real time:

● Tokengram: 10.0k saved · 73%  ⚡+7.4k now

Hover for the full breakdown (search, grep, read-block, bash inspect, today's totals). Click the bar for the action menu — toggle enforcement, refresh, reindex, set license, reset history.

Privacy

Tokengram is fully local. No code, no queries, no embeddings ever leave your machine. The only network call is one-time license validation (cached for 7 days offline).

License

Commercial. See LICENSE.txt. $2/month subscription required for activation after a 7-day trial.

Support

Issues: github.com/kinleee/tokengram/issues
Email: n.ilkin.humbatov@gmail.com

Built by Ilkin Humbatov.