How to give Open WebUI better memory

What Open WebUI documents for memory

The docs put Knowledge Bases at the center: they let you ask questions across PDFs, documents, and other files in Open WebUI (Open WebUI docs). Two building blocks are described. Knowledge retrieves relevant passages on demand. Notes inject full content into every message (Open WebUI docs).

Knowledge is aimed at document sets that would exceed the model's context window if sent in full (Open WebUI docs). For a wider view of this approach, see RAG vs a personal knowledge base.

Attach a knowledge base to a model

You attach a knowledge base to a model in Workspace > Models > Edit, then start a conversation with that model (Open WebUI docs). Attaching scopes the model to only those documents. Leaving it unscoped lets the model discover everything the user has access to (Open WebUI docs).

Choose a retrieval mode

Each collection offers two modes: Focused Retrieval (RAG) to let the AI search large collections efficiently, or Entire Document to inject a whole document word-for-word when precision matters (Open WebUI docs). Focused Retrieval uses RAG to find and inject the most relevant chunks based on the user's query (Open WebUI docs).

In Full Context mode, the complete content of the file is injected into every message, with no chunking and no semantic search (Open WebUI docs).

Hybrid search and agentic retrieval

Hybrid search combines BM25 keyword search with vector search and cross-encoder reranking (Open WebUI docs). With agentic retrieval, models browse, search, and read your documents autonomously (Open WebUI docs).

Watch the Full Text Threshold

Files and text attachments up to the Full Text Threshold setting in Admin Settings > Documents are included in full without choosing a special mode (Open WebUI docs). Knowledge base attachments are not affected by the threshold (Open WebUI docs).

Add an outside knowledge base over MCP

Open WebUI supports MCP natively starting in v0.6.31 (Open WebUI docs). Only administrators can add MCP servers, under Settings > Admin > Integrations, and the Type must be set to MCP (Streamable HTTP), not OpenAPI (Open WebUI docs). Open WebUI's MCP support is Streamable HTTP only (Open WebUI docs).

An admin adds the server once and scopes it with Access Control to the right users or groups (Open WebUI docs). Bearer mode is only for servers that require a specific API token, and the Key field must be populated (Open WebUI docs). One caution from the docs: OAuth-connected MCP tools break on every container restart unless the required environment variable is set in your Docker setup (Open WebUI docs).

Where KnowMan fits

KnowMan runs a hosted remote MCP server at https://api.knowman.io/mcp over Streamable HTTP (knowman-mcp), which matches what Open WebUI requires. It accepts OAuth sign-in or an API key sent as Authorization: Bearer <key>, and exposes six tools: kb_search, kb_ask, kb_get_item, kb_save, kb_recent, and kb_memories (knowman-mcp).

KnowMan also ships a local stdio server, npx -y @knowman/mcp (knowman-mcp). Open WebUI's documented MCP support is Streamable HTTP only (Open WebUI docs), so the hosted server is the path the docs describe.

The split of work is simple. Open WebUI knowledge bases hold files you upload (Open WebUI docs). KnowMan holds what you read over time: saved articles and links, read and tagged, with cited answers, plus claims, decisions, and playbooks you keep (knowman-mcp). See give your AI context for the general idea.