I Tested 27 Claude Code Runs: Does Graphify Actually Save Tokens?

Claude Code memory, tested: does Claude Code re-scan your codebase every session, and does Graphify really save tokens? I set up one real repo three ways (plain Claude Code, a CLAUDE.md from /init, and Graphify 0.9.86), asked the same three questions three times each and counted every token across all 27 runs. ---- Higgsfield AI — Create AI Videos & Visuals Get 10% OFF here: https://higgsfield.ai?fpr=thomas-dccf7e 🚀 DYNAMOUS AI COMMUNITY Want to learn agentic coding with live daily events and workshops? Check out Dynamous AI: https://dynamous.ai/?code=646a60 Get 10% off here 👉 https://shorturl.smartcode.diy/dynamous_ai_10_percent_discount ⚡ HOSTINGER — RELIABLE HOSTING FOR YOUR PROJECTS (10% OFF) Whether you're shipping a portfolio, a side project, n8n flows, or AI agents — I use Hostinger for fast, affordable VPS + web hosting. Get 10% off here 👉 https://hostinger.com/DIYSMARTCODE n8n: https://n8n.partnerlinks.io/diysmartcode (Affiliate links — costs you nothing, supports the channel.) ---- Chapters 0:00 Claude Code Memory Tested: Does Graphify Really Save Tokens? 0:35 Does Claude Code Re-Scan Your Codebase Every Session? (/context) 1:45 Where Claude Code Tokens Really Go: Prompt Caching on One Question 2:30 CLAUDE.md and /init: Does Built-In Memory Cut Token Usage? 3:29 Dynamous AI Community (Werbung) 4:04 Graphify Benchmark vs 9 Real Runs: Fewer Tokens or More? 5:39 Claude Code Auto Memory, CLAUDE.md and --resume Across Sessions 6:54 When a Code Graph or an MCP Index Actually Helps 7:52 Verdict: Claude Code Token Optimization That Works What you will see in this 9-minute Claude Code token test: - The claim from a reel with 1.3M views (Claude re-reads your entire codebase every session) checked with /context in a fresh session: 27.3K tokens before the first prompt, none of them from the repo - The test repo, Textual: about 2,100 files and an estimated 2.65M tokens of text, more than the 1M context window Opus has in Claude Code - Where a question's tokens go: 82% were cache reads on the median plain run, billed at 5% of the input price on the API - /init and CLAUDE.md: 47 seconds once, 43 lines, about 1K tokens per session, and no token savings on this repo (115.7K vs 125.6K per question, inside a 61K run-to-run swing) - Graphify 0.9.86: graph built locally in 27 seconds, its own 9.0x benchmark measured against pasting the whole corpus, and 9 of 9 real runs going back to grep (124.3K per question, about 2 minutes instead of 40 seconds on the trace) - Across sessions: auto memory (first 200 lines or 25KB), what it deliberately skips, and why a resumed session cost twice as much after the one-hour cache expired - Where graphs and indexes do help: Graphify's own coverage benchmark on a roughly million-line repo, MCP code search and code intelligence plugins - Why even a real 70% saving would not turn Pro into Max 5x - reel: https://www.instagram.com/reel/DYnPUfru7aw/ - Test repo, Textual at commit 5e5b7ef: https://github.com/Textualize/textual - Graphify: https://github.com/Graphify-Labs/graphify - Graphify benchmarks: https://github.com/Graphify-Labs/graphify/blob/v8/BENCHMARKS.md - How Claude Code works in large codebases (Anthropic): https://claude.com/blog/how-claude-code-works-in-large-codebases-best-practices-and-where-to-start - Claude Code memory, CLAUDE.md and auto memory: https://code.claude.com/docs/en/memory - Prompt caching in Claude Code: https://code.claude.com/docs/en/prompt-caching - Manage costs effectively: https://code.claude.com/docs/en/costs - Claude Code in large codebases: https://code.claude.com/docs/en/large-codebases - Claude API pricing: https://platform.claude.com/docs/en/about-claude/pricing - What is the Max plan: https://support.claude.com/en/articles/11049741-what-is-the-max-plan Code graph, or plain search with a short CLAUDE.md? Which one runs in your repo? #ClaudeCode #Graphify #ClaudeCodeMemory #ClaudeCodeTokens #TokenOptimization #ClaudeTokenUsage #ClaudeMemory #CLAUDEmd #PromptCaching #ContextWindow #AutoMemory #ClaudeAI #Anthropic #CodeGraph #KnowledgeGraph #MCP #ClaudeCodeTutorial #AICoding #AgenticCoding #AItools #VibeCoding