Open WebUI Memory & Personalization

0Orientation & objectives

Open WebUI includes a memory system that stores facts, preferences, and context about a user and carries them across separate conversations. With the introduction of Native Tool Calling, that system changed from a passive injection mechanism into an active, model-managed long-term memory. This course covers what the memory system stores, how quality models read and write it through eight built-in tools, how memories are typed and grouped, how administrators enable and constrain the feature, and how memory data is scoped and protected.

Learning objectives. On completion a participant will be able to describe how Open WebUI memory operates in both manual and native modes, identify the eight memory tools and the two memory types, enable the feature at the model and administrative levels, configure background review and injection budgets, and state how memory data is stored and scoped.

Format	Self-paced; approximately 20 minutes
Formative questions	8 multiple-choice, distributed (feedback, not scored)
Final assessment	8 multiple-choice questions; pass mark 75%
Sources	Open WebUI documentation, Memory & Personalization and Environment Variable Configuration (captured 2026-07-08)

Source: docs.openwebui.com — Memory & Personalization; env-variable cross-references from Environment Variable Configuration.

1What memory is, and how it works

Learning objective. Distinguish the two ways memories are used, and state the documented dependence of memory quality on the model.

The memory system stores snippets of information about the user — for example, "I prefer Python for backend tasks" or "I live in Vienna." There are two ways these memories are used:

Manual management. Users add, edit, or delete memories at Settings > Personalization > Memory.
Native memory tools (Agentic Mode). With Native Function Calling enabled, quality models manage memory autonomously through built-in tools.

Model-dependent quality. The documentation states that how well memories are stored and recalled depends heavily on the model: frontier models manage memory well, while small local models may store or retrieve information inconsistently. Autonomous management works best with frontier models (the documentation names GPT-5, Claude 4.5+, Gemini 3+).

The upgrade introduced with Native Tool Calling is the shift from a passive injection mechanism to an active, model-managed long-term memory: the model itself decides when to store and retrieve, rather than the system injecting a fixed block of remembered text.

Check your understanding

1.1 Where do users manually add, edit, or delete memories?

1.2 The documentation says memory quality depends heavily on —