A lightweight plugin that gives your local LLM persistent memory across sessions. Reads from and writes to a plain markdown file on your machine, allowing the model to remember names, preferences, decisions, and other details between conversations. Three tools: read_memory, save_memory, and update_memory.
A lightweight persistent memory plugin for LM Studio. Gives your local LLM the ability to remember facts, preferences, and context across sessions using a plain markdown file stored on your machine.
What it does
Reads memory on session start — the model loads your memory file before responding to anything
Saves facts as you go — when you share your name, location, preferences, decisions, or any detail worth keeping, the model writes it immediately
Updates existing entries — corrections replace old information rather than duplicating it
Tools
read_memory — reads the memory file at session start
save_memory(fact) — appends a new dated entry
update_memory(old_text, new_text) — finds and replaces an existing entry
Setup
Install the plugin and enable it in LM Studio
In the plugin settings, set Memory File Path to the full path where you want your memory file stored. Example:
The file and folder will be created automatically on first use.
Set all three tools to Always Allow so memory reads and writes happen without confirmation prompts.
System prompt
For reliable automatic memory behavior, instruct your model to call read_memory at the start of every session and save_memory whenever new information worth retaining is shared. The exact wording will depend on your model and preferences.
Notes
The memory file is plain markdown — you can read and edit it directly at any time
Tested with Qwen 3.5 in LM Studio 0.4.9
Model behavior depends on how well the model follows tool-use instructions — larger models are more reliable