How to teach LM Studio your knowledge

What LM Studio means by knowledge

LM Studio says chats do not train or otherwise change what a model knows. The model uses content in the chat or configuration, such as a system prompt. LM Studio explains chat context.

A model's context is its working memory and has a maximum size. Short files that fit can be added in full, while very long documents may use retrieval to provide relevant portions. LM Studio describes context and RAG.

Set reusable instructions with a Preset

Create a Preset with a system prompt for instructions you want to reuse across chats. A Preset can also store settings such as Temperature, Top P, and Max Tokens, and you can switch presets for different uses. See LM Studio Presets.

Presets are JSON files. LM Studio documents importing them from disk, pulling them from a URL, or publishing them to the LM Studio Hub. See the Presets guide.

Attach files to provide source material

Attach a .docx, .pdf, or .txt file to a chat to give the model additional context. If a document fits within the model's context, LM Studio adds its contents in full. Read the LM Studio file and RAG guide.

For very long files or several long files, LM Studio can use Retrieval-Augmented Generation to retrieve relevant portions. Its documentation notes that this can require tuning and experimentation. Queries may work better when they include terms and ideas likely to appear in the source material. See LM Studio's RAG guidance.

Connect an MCP server

LM Studio supports local and remote MCP servers starting with version 0.3.17. Its documentation describes adding servers by editing mcp.json or using an Add to LM Studio button when available. See LM Studio MCP setup.

A remote server entry can include a URL and authorization headers. MCP servers may run code, access local files, or use the network, so use sources you trust. Some servers may also consume enough tokens to overflow a local model's context. Review LM Studio's MCP guidance.

Add KnowMan as a hosted MCP server

Add https://api.knowman.io/mcp as a remote MCP server. KnowMan's hosted server supports OAuth approval or an API key sent as an Authorization Bearer header. See the KnowMan MCP documentation.

For a client configuration that accepts remote server URLs and headers, the documented shape is:

```json { "mcpServers": { "knowman": { "url": "https://api.knowman.io/mcp", "headers": { "Authorization": "Bearer <key>" } } } } ```

LM Studio documents URL and authorization-header fields for remote MCP entries. KnowMan's OAuth consent separates read access for search, reading, and asking from write access for saving links. LM Studio MCP setup · KnowMan MCP documentation.

Ask about your saved reading

After connecting KnowMan, ask the assistant to search your saved knowledge or save a URL. KnowMan's MCP tools include searching, asking questions, reading an item, saving a link, viewing recent items, and listing memories. See the KnowMan MCP tools.

The hosted server's documented OAuth flow opens a KnowMan consent page for approval. It also accepts an API key in the Authorization Bearer header. Read the KnowMan MCP setup.

Where KnowMan fits

Presets keep reusable instructions, and attached files give a chat source material. KnowMan adds a connected library that an assistant can search and ask about through MCP. LM Studio Presets · LM Studio file context · KnowMan MCP.

For more on connecting an assistant to saved knowledge, see connect any AI to your notes and MCP server knowledge bases.