The video provides a highly specific, actionable technical tip that solves a common pain point (token limits) for a niche audience of developers using a popular tool.
Summary
The creator explains how to save tokens when using Claude Code by utilizing the /init command. This command generates a CLAUDE.md file that provides the necessary project context to the AI, preventing it from re-reading the entire codebase in every new session.
Structure
1Identify the problem of high token usage in Claude Code.
2Explain why Claude re-reads the entire codebase every session.
3Introduce the /init command as a solution.
4Demonstrate how /init creates a CLAUDE.md file.
5Explain how this file provides context to Claude without re-reading all files.
Product placement
Claude Code is the subject of the video, demonstrated throughout the screen recording.
Original caption
Stop burning Claude Code tokens. Use the /init command to auto-generate a CLAUDE.md file that gives Claude the context it needs — so you're not re-explaining your project every single session. Work smarter, spend less. #claudecode #aiproductivity #tokensavingtips #claudeai
More from @kamranaiinsights
Their other posts in the index, biggest breakout first.
On-screen text
Save Claude Tokens
/init
Transcript
If you are using Claude Code, then you are probably using a lot of tokens. And sab ko yahi masla rehta hai ki yaar hamara Claude Code ka daily usage session hai wo bohot jaldi over ho jata hai. So, main aapko ek simple sa hack bataunga. Sabse pehle problem ko describe karta hoon ki kaunsi aisi cheez hai jo ki basically hamare tokens khaye ja rahi hai lekin humein idea hi nahi hai. Toh suppose karo ki aapne koi bhi jo hai wo ek project build kiya hua hai. Aap ek session ke upar kaam kar rahe the, aapne jo hai wo close kar diya. Agle din jab aap aaoge, aap new session create karoge, theek hai, Claude ko dobara launch karoge and usko koi bhi prompt doge ki yaar ab maine jo hai wo ye kaam karna hai ya mere liye ye feature implement karo, toh wo kya karega? Wo aapka poora jo code hoga previous, usko dobara se read karega. Kyunki usko idea ho paye ki humne previous day mein kya build kiya tha aur actual mein hamara project kya hai. Theek hai na? Toh wo kya karega ki wo poora sa aapka code jo hai wo dobara se read karega aur read karne ke andar wo aapke bohot zyada tokens ko jo hai wo pehle hi consume kar lega. Theek hai? Abhi toh ye maine jo hai wo chaar-paanch fileon ka code hai, lekin suppose karo aapki jo hai wo 100 files hain aur bohot bada code hai, bohot bada project aap build kar rahe ho, theek hai? Toh agar aap agle din aaoge, agle din aa ke aap usko kahoge ki yaar maine ye feature implement karna hai, toh wo feature implement karne se pehle wo apna context lega ki yaar maine actual mein build kya kiya hai. Theek hai? Toh wo kya karega ki wo aapki poori file ko again se read karega aur jab again se read karega 100 files ko, toh aapka saara token saara usage jo hai wo pehle hi consume ho jayega. Theek hai? Jab aapne ye feature implement kiya, toh aap kehte ho ki yaar ye toh problem ho gayi. Toh simple sa hack ye hai ki jab bhi aap daily session complete karo, jab bhi aapne daily ka kaam kar liya, toh aapne ek command run karni hai. Jis tarah se main simply aapko ek bata deta hoon. Aapne kya karna hai ki aapne apne Claude Code ke session ke andar simply yahan pe ja ke search karna hai slash aur yahan pe aapne jo hai wo init command ek run kar deni hai. Theek hai? Init, ye wali command hai. Init. Main just simply isko jo hai wo run kar deta hoon. Ab ye kya karega ki jo bhi session ke andar aap init command run karenge, us session tak ye aapke poore code base ko ek baar read kar lega aur read karne ke baad aapke liye ek MD file create kar dega. Means ki ek aisa data base, ek aisi file create kar dega jiske andar aapke poore code base ki saari information hogi. Ab wo dekhne mein tum jab create hogi, main aapko tab dikhaunga, fayda kya hoga ki agli baar aap suppose karo ki aapne jab new session create kiya ya agle din aaoge aap, aap jab aa ke usko kahoge ki yaar, ki ya maine ye feature implement karna hai, this and that, kuch bhi aap baat karoge, toh wo aapka poora code base read nahi karega. Wo aapki just ek ye file implement karega, CLAUDE.md, usko read karega. Ab jis tarah se main aapko bataun, isne jo hai wo file jo hai wo create kar di hai. Toh main just simply isko open kar deta hoon. Toh is agar main is file ko agar aapko thoda sa dikhaun, toh is file ke andar isne jo hai wo mere project ki saari details jo hai wo add kar di hain. Toh fayda kya hoga ki jab bhi aap new session ke andar aaoge aur usko koi bhi question karoge ki yaar, Claude mere liye ye feature build karo, this and that, kuch bhi aap usse bologe, toh wo ab aapka poora code read nahi karega. Wo just aapki ye file, CLAUDE.md, jo hai wo isko read karega aur isko read karne ke baad uske paas proper context ho jayega. Then wo aapke liye features implement karega aur aap carry on karoge. Ab suppose karo ek taraf, ek taraf wo poora code base read kar raha hai aur ek taraf just wo ek ye file read kar raha hai. File read karne ke baad aapko easily jo hai wo implement Claude Code ko ho jayega ki yaar haan, ye project is se related hai aur maine isko is wajah se jo hai wo kaam karna hai aage. So, this is the trick, jo mujhe lagta hai ki agar hum ye tarah se implement karein, toh hum Claude Code ke tokens ko kam consume kar sakte hain.
