Skip to content

Choose Your Path

You do not need every memtomem feature at once. Pick the outcome you need now, finish one working path, and add another layer only when it becomes useful.

I want to…Start withYou are done when…
Prove that memory storage and search workQuick Startmm search returns the memory you just added
Recall a decision in a new AI sessionMemory Across Sessionsa new session returns the saved reason and source
Connect Claude Code, Codex, or another MCP clientConnect an AI Clientthe client can call mem_status
Search existing notes, docs, code, or built-in AI memoryIndex and Import Existing Contenta known phrase returns with its source path
Add response compression or proactive memory surfacingAdd STM to an MCP Serverone proxied call appears in mms stats --source mcp
Separate personal, agent, and shared memoryMulti-Agent Collaborationone agent shares a reviewed memory and another retrieves it
Maintain skills and commands across runtimesContext Gatewaya previewed Store version is pushed to a selected runtime
Check local storage and external-service boundariesLocal-First & Privacyyou know which enabled integrations can send data off-device
Diagnose or operate an existing installationTroubleshooting and Operations & APIhealth checks pass or point to a specific recovery step

Start with LTM (memtomem). It stores durable memory and searches it on demand. Add STM (memtomem-stm) only when an MCP workflow needs response compression, caching, or proactive memory surfacing.

NeedLTMSTM
Save a decision across sessionsYesNo
Search notes, docs, and codeYesNo
Share one local store across AI clientsYesNo
Compress an MCP tool responseNoYes
Attach relevant LTM memory to a proxied callProvides the memoryPerforms the surfacing
Required for the first successful setupYesNo

STM does not replace LTM and does not write every proxied response back into durable memory. The two packages run independently and communicate through MCP when surfacing is enabled.

Installing a plugin or registering the MCP server does not automatically:

  • index your whole project;
  • watch every file in the background;
  • import all Claude Code, Codex, or Gemini/Antigravity memory;
  • save entire conversations or every tool response;
  • synchronize databases between different computers.

Use an explicit indexing, import, save, or synchronization workflow when you want one of those outcomes.

  1. Complete Quick Start without an MCP client.
  2. Connect one AI client and verify mem_status.
  3. Save one confirmed decision and retrieve it in a new session.
  4. Index one small, non-sensitive docs directory and verify a source-backed result.
  5. Add STM only if you can name the MCP server whose responses you want to proxy.

Keep permanent project rules in CLAUDE.md, AGENTS.md, or the equivalent runtime instruction file. Put decisions, findings, and repeatable procedures in memtomem when you want to retrieve them later.

Task guides recommend one safe path. The reference pages retain the full released command and configuration surfaces: