Your AI agent forgets who you are every session. Give it a memory that actually learns.Most "agent memory" is an opaque vector blob you can't read, locked to one vendor, and it never improves — it just stores. The Agent Memory Kit is the opposite: a folder of plain Markdown your agent reads, you can edit, git tracks, and a tiny script keeps clean. It learns from your corrections instead of just storing them.No database. No server. No subscription. Python standard library + a folder.WHAT IT DOES — four moves, distilled from the author's own working self-improving memory system:• CAPTURE — save durable facts and corrections as provenance-tagged notes• RECALL — your agent checks memory before tasks; recall ranks notes by keyword and surfaces any [[wikilinks]] they contain• CURATE — weekly cleanup: flags exact-duplicate notes and archives thin agent-written ones so memory stays sharp (it NEVER deletes — it archives, recoverable)• PROFILE — a 15-question personality quiz turns into a finished CLAUDE.md that knows who you are, how you work, and what you never want to see againTHE MATURITY LADDER (start where you are): L1 paste-a-profile → L2 capture → L3 recall → L4 curate → L5 self-improving loop. Most people stop at a static CLAUDE.md. Levels 2-5 are the moat — and the kit walks you up every rung.ONE-PASTE SETUP: a Drop-In MD you paste into your agent runs the entire setup as guided steps. You answer the quiz, it builds your profile, scaffolds your memory folder, and wires the habits. Done.THE HONEST COMPARISON: we researched mem0, Letta (MemGPT), Zep, vector RAG, and Claude's native memory — and we show you exactly where this wins (zero-dependency local, human-readable + git-able, self-improving, provenance + archive-never-delete, portable across agents, free) AND where it doesn't (million-scale search, hosted compliance certs). We'd rather tell you the truth than oversell.ALREADY USE OBSIDIAN? Point the tool at a folder inside your vault — it keeps its notes there, so they live in your vault and Obsidian graphs the [[wikilinks]] recall surfaces. Knowledge-graph retrieval (Graphify, a separate third-party tool) is an optional power upgrade. Don't use them? The folder works great without either.WHAT YOU GET• memory.py — the portable tool (capture/recall/curate/status), stdlib-only, offline• The 15-question Personality Quiz → a finished CLAUDE.md• The Drop-In Setup MD (one paste = full setup)• The Build Playbook with the L1→L5 ladder• An illustrated setup guide + the full honest comparison• A free quiz you can share, installer + clean uninstaller (your memory survives)• Claude Code first; AGENTS.md / soul.md = same pattern, rename the fileREQUIREMENTS (please read before buying)• An AI coding agent (Claude Code, Codex, Cursor, etc.)• macOS or Linux — Windows works via WSL• Python 3.8+ (already installed on Mac/Linux)• No API keys required for the core; Obsidian / Graphify are optional power-upsThis portable kit is the transferable core of a larger, stack-specific system — not that full system. Comparison figures describe third-party systems as of mid-2026 and change; verify with each vendor. Not affiliated with, endorsed by, or sponsored by Anthropic, mem0, Letta, or Zep. Claude is a trademark of Anthropic PBC; other names belong to their owners.