Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/mimi180306/claude-persistent-memory/claude-mdgit clone --depth 1 https://github.com/MIMI180306/claude-persistent-memoryWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/mimi180306/claude-persistent-memory/claude-md)<a href="https://agentmods.dev/instructions/mimi180306/claude-persistent-memory/claude-md"><img src="https://agentmods.dev/badge/instructions/mimi180306/claude-persistent-memory/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00867 | $0.00867 |
| Opus 5 | $0.00434 | $0.00434 |
| Sonnet 5 | $0.00173 | $0.00173 |
| Haiku 4.5 | $0.00087 | $0.00087 |
Grade A, and why
claude-persistent-memory CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 66 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
What This Is
A persistent memory system for Claude Code — hybrid BM25 + vector semantic search with LLM-driven structuring and automatic clustering. It gives Claude Code sessions long-term memory that persists across conversations.
Commands
npm run embedding-server # Start embedding service (TCP :23811, loads bge-m3 ~2GB RAM)
npm run llm-server # Start LLM proxy service (TCP :23812, requires Azure OpenAI)
npm run mcp-server # Start MCP server (stdio, used by Claude Code)
npm run rebuild-vectors # Rebuild all embeddings (after model change)
Both TCP servers must be running for full functionality. The MCP server is configured in .mcp.json and launched by Claude Code automatically. Hooks are configured in .claude/settings.json.
Architecture
The system has two retrieval channels:
Pull channel — MCP server (services/memory-mcp-server.js) exposes 4 tools (memory_search, memory_save, memory_validate, memory_stats) that Claude invokes on demand via stdio.
Push channel — 5 hooks auto-inject memory context into Claude's conversation:
UserPromptSubmit→ embeds user query, searches, prepends results to prompt via stdoutPreToolUse→ searches on Edit/Write/Bash tool context, injects viaadditionalContextPostToolUse→ searches on tool context + result, injects viaadditionalContextPreCompact→ spawnscompact-analyzer.jsin background (non-blocking), which sends full transcript to LLM for memory extractionSessionEnd→ incremental transcript analysis + auto-clustering + mature cluster merging
Internal service communication
Hooks and MCP server communicate with the two TCP servers via lib/embedding-client.js and lib/llm-client.js. Both clients use raw TCP sockets with newline-delimited JSON protocol.
Data layer
lib/memory-db.js is the core module — manages SQLite database with three virtual tables:
memories— main table with structured content, confidence scores, cluster assignmentsmemories_fts— FTS5 full-text search (BM25 ranking, synced via triggers)memories_vec— sqlite-vec cosine similarity (1024-dim bge-m3 embeddings)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 66 lines · 867 tokens per session scan A 0cc6fde0f1af
claude-persistent-memory CLAUDE.md is an instructions file published in the GitHub repository MIMI180306/claude-persistent-memory (8 stars, last pushed 6mo ago), licensed MIT. It adds 867 tokens to every session, about $0.0043 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
wife AGENTS.md
Instructions for ma-nucho-pro/wife, covering if the user wants to install it, if the user wants to change something and rules that are not negotiable.
mnemo-mcp AGENTS.md
AGENTS.md instructions for sshisto/mnemo-mcp, covering agents.md - mnemo-mcp, build / lint / test commands, run a single test file, run a single test class and run a single test function.
claude-memory-kit CLAUDE.md
Instructions for awrshift/claude-memory-kit, covering working on memory kit, rules that are easy to get wrong, before committing a change to hooks or skills, docs that must stay true in the same commit and language.
muninn CLAUDE.md
Claude Code instructions for ravnltd/muninn, covering claude.md — muninn memory system, what happens automatically (no action needed), the 4 mcp tools (manual escape hatch), architecture and install / update.
wife CLAUDE.md
Instructions for ma-nucho-pro/wife, covering wife — notes for the agent reading this, if the user wants to install it, if the user wants to change something and rules that are not negotiable.
claude-memory-template CLAUDE.md
Instructions for lukasz-fedor/claude-memory-template, covering my project, session continuity, work conventions, anti-hallucination protocol and git workflow.