How to give LibreChat better memory

What LibreChat memory stores

LibreChat memory is a small store of facts about you, saved as key/value pairs, that is added to the context of every new message (LibreChat docs). It is not a search over your chat history. Nothing from old conversations is recalled unless it was saved as a memory entry.

The whole store goes into each request, so it is meant for a handful of durable facts, not notes or documents. Better memory starts with that split. Short facts about you stay in memory, and everything else moves to prompts, file features, or a knowledge base.

Keep memory entries small and deliberate

When Memory is on, the model is instructed to use the memory tools only when you ask it to remember, update, or forget something, so say it plainly, for example: remember that I prefer short answers (LibreChat docs).

Keys use lowercase letters and underscores only, and each key can exist once. Once the store is full, new saves fail with "Memory Storage Full" until you delete something, and individual entries also have a length cap (LibreChat docs). Review old keys before adding new ones.

Agents, separate stores, and admin controls

In chats with an agent, the composer palette has no Memory option. The agent uses memory if Memory is enabled in the Agent Builder (LibreChat docs).

An agent can be set to keep memories separate, so it reads and writes its own store, isolated from your shared memories and from other agents. Some deployments also run a background memory agent that reviews each exchange and saves facts on its own, without you asking. Whether that happens is your admin's choice (LibreChat docs).

Turning memory off stops LibreChat from adding saved memories and stops offering the Memory tool, and your entries are kept and come back when you turn it on again. The opt-out toggle is only shown when the admin allows users to opt out, and memory can also be limited by role (LibreChat docs).

Put stable knowledge in agent instructions and files

An agent is a saved setup: a model, a system prompt, and the tools and files it can use (LibreChat docs). Stable preferences belong in that system prompt, and an agent's instructions can link to a saved prompt instead of inline text, which keeps long instructions out of the chat window.

Files added under File Context have their text extracted once, at upload, and that text goes into the agent's context on every message. Because the text rides along with every message, large files there cost tokens on every turn.

Use File Search for large documents

File Search turns the files you attach into a searchable knowledge base. The agent pulls the most relevant passages when it needs them, so it copes with large document sets (LibreChat docs). Long documents belong here, or in a dedicated library, not in memory entries.

For the difference between retrieval over attached files and a curated personal knowledge base, see RAG vs a personal knowledge base.

Turn on MCP servers in LibreChat

MCP is an open standard for exposing tools to a model, and LibreChat can use any server that speaks the protocol (LibreChat docs). Click a server to turn it on for the conversation. It shows up as a chip on the composer, and every tool the server exposes becomes available to the model. Pick a model that supports tool calls, or the model will never call the server's tools.

Connections are per user, saved credentials are never used for anyone else, and an idle connection is closed after 15 minutes by default and reopened the next time you use the server (LibreChat docs).

KnowMan's MCP server

KnowMan offers a hosted remote MCP server at https://api.knowman.io/mcp (Streamable HTTP). Sign in with OAuth, which opens a KnowMan consent page for you to approve, or send an API key as Authorization: Bearer <key> (KnowMan MCP). For clients with MCP OAuth, the documented setup is to add the server URL, sign in, and approve.

The server exposes six tools: kb_search, kb_ask, kb_get_item, kb_save, kb_recent, and kb_memories (KnowMan MCP). KnowMan also runs locally as a stdio server started with npx -y @knowman/mcp, on Node 22 or newer. For the wider pattern, see an MCP server knowledge base.

Where KnowMan fits

LibreChat memory holds durable facts about you, File Context sends file text with every message, and File Search works on files you attach by hand (LibreChat memory docs, agents docs). None of these is a library that grows as you read the web.

KnowMan saves articles and links, reads and tags them, answers questions with citations, and keeps memories such as claims, decisions, and playbooks. Over MCP you can ask your chat to "search my KnowMan for ..." or "save this URL" (KnowMan MCP). For the broader idea, see persistent memory for AI agents and give your AI context.