What LM Studio means by context
LM Studio calls context the working memory of an LLM and states that it has a maximum size (LM Studio docs). Context is measured in tokens, and one token is often about three quarters of a word (LM Studio docs).
What the model remembers between chats
LM Studio is direct here: the model does not learn from chats, and it only knows the content present in the chat or provided through configuration options such as the system prompt (LM Studio docs).
So better memory is not a toggle. It is context you re-supply each time, or a tool the model can call to fetch what it needs. The rest of this page covers both.
Give a chat more context with attached files
You can attach document files, .docx, .pdf, and .txt, to chat sessions in LM Studio, which gives the model additional context (LM Studio docs). If a document is short enough to fit the model's context, LM Studio adds the file contents to the conversation in full (LM Studio docs). For quick questions over a single paper or report, this is often all you need.
Get more out of long documents
When a document is very long, LM Studio opts into Retrieval Augmented Generation, or RAG (LM Studio docs). RAG means pulling relevant bits out of one very long document, or several documents, and providing them to the model for reference (LM Studio docs).
The docs offer one tip: provide as much context as possible in your query, and mention the terms and ideas you expect to find in the source material (LM Studio docs).
Keep standing instructions in Presets
Presets bundle a system prompt and other parameters into one configuration that is reused across chats (LM Studio docs). Every parameter under the Advanced Configuration sidebar can be recorded in a named Preset, and LM Studio suggests using Presets to build your own prompt library (LM Studio docs).
Presets are stored on disk, in ~/.lmstudio/config-presets on macOS or Linux and %USERPROFILE%\.lmstudio\config-presets on Windows, and you can import them from file or URL or publish them to the LM Studio Hub (LM Studio docs).
Add an MCP server in LM Studio
Starting with version 0.3.17, LM Studio acts as an MCP Host, so you can connect MCP servers to the app and make them available to your models, and both local and remote servers are supported (LM Studio docs).
To add one, switch to the Program tab in the right hand sidebar and click Install, then Edit mcp.json (LM Studio docs). LM Studio follows Cursor's mcp.json notation (LM Studio docs).
Two cautions from the docs: never install MCP servers from untrusted sources, and watch token use, because some servers can bog down a local model and trigger frequent context overflows (LM Studio docs).
The documented KnowMan MCP setup
KnowMan runs an MCP server two ways. Hosted at https://api.knowman.io/mcp over Streamable HTTP, reached by OAuth sign in or an Authorization Bearer API key, or locally as a stdio server from the npm package @knowman/mcp, started with npx -y @knowman/mcp on Node 22 or newer (KnowMan MCP).
The server exposes six tools: kb_search, kb_ask, kb_get_item, kb_save, kb_recent, and the read-only kb_memories (KnowMan MCP). OAuth grants knowman.read for search, read, and ask, and knowman.write for saving links, and access tokens last one hour and refresh automatically (KnowMan MCP).
Setup needs one KnowMan account. For clients with MCP OAuth, you add the server URL, sign in, and approve (KnowMan MCP). KnowMan also documents a Cursor mcp.json entry for the hosted server with a Bearer key header, matching the mcp.json format LM Studio supports (KnowMan MCP, LM Studio docs).
Where KnowMan fits
Attached files give the model additional context, RAG pulls relevant bits from long documents, and Presets bundle a system prompt and parameters for reuse across chats (LM Studio docs, LM Studio docs). These cover a lot of ground, and they are different from a library that grows as you read.
KnowMan saves articles and links, reads and tags them, answers questions with citations, and keeps memories such as claims, decisions, and playbooks (KnowMan). LM Studio supports remote MCP servers (LM Studio docs), and KnowMan documents an mcp.json entry for its hosted server that matches that notation (KnowMan MCP). Mind the token caution above about local models (LM Studio docs).
For the wider pattern, see a knowledge base for AI agents, persistent memory for AI agents, and give your AI context.
Questions people ask
Does LM Studio remember previous chats?
No. LM Studio states that the model does not learn from chats and only knows the content present in the chat or provided through configuration options such as the system prompt (LM Studio docs). Starting with version 0.3.17, LM Studio acts as an MCP Host, so you can connect MCP servers to your models (LM Studio docs).
How do I add documents to an LM Studio chat?
Attach .docx, .pdf, or .txt files to the chat session, which gives the model additional context (LM Studio docs). Short documents are added to the conversation in full, and very long ones go through RAG (LM Studio docs).
What are Presets in LM Studio?
A Preset bundles a system prompt and other parameters into one configuration you can reuse across chats (LM Studio docs). They are stored on disk, and you can import them from file or URL or publish them to the LM Studio Hub (LM Studio docs).
Can LM Studio use MCP servers?
Yes. Starting with version 0.3.17, LM Studio acts as an MCP Host and supports both local and remote servers (LM Studio docs). You add them by editing mcp.json from the Program tab in the right hand sidebar (LM Studio docs).