The question and answer posted here here was eye opening, so I started to mess around with disabling the Citations Capability and was amazed at how effective this is at improving the Input caching rates for my requests (went from like 1-10% to sometimes as much as 50% or greater). This is going to be a huge savings as we roll out OWUI to our organization.

My question surrounds the Workspace and Base Model features in OWUI:

My setup / intended configuration:

• A connected base model with a system prompt configured in Open WebUI.
• A custom model created under Workspace → Models, selecting that model as its Base Model (From).
• The workspace model has its own system prompt for its specialized role.

The docs describe workspace models as wrappers with their own system prompts, but I haven’t found an explicit explanation of whether the base model’s configured prompt is also included when using the workspace model system prompt.

My practical goal is to apply shared instructions for reusing/caching web research and producing grounded citations, while keeping each workspace model’s role-specific instructions separate. For example:

• Reuse previously retrieved research when appropriate rather than repeating searches unnecessarily.
• Fetch source pages when needed to support citations, rather than treating search snippets as verified page content.
• Cite the sources actually used and avoid fabricated or mismatched citations.

The closest upstream reference I was able to find in the owui docs was Chat Parameters, particularly its system-prompt hierarchy section:

1. System-prompt hierarchy

Open WebUI — Chat Parameters: System Prompt and Advanced Parameters Hierarchy Chart

It documents three levels:

• Per-account: Settings → General → System Prompt.
• Per-chat: Chat Controls → System Prompt.
• Per-model: Workspace → Models → System Prompt.

The page says the per-model system prompt takes precedence and cannot be overridden by a user’s account-level or chat-level prompt. It does not distinguish a base model’s Open WebUI-configured prompt from a derived workspace model’s prompt.

2. Workspace models and their system prompts

Open WebUI — Models
System prompt and variables section

This describes workspace models as configuration presets layered over a selected base model, with their own system prompts, knowledge, tools, and parameter overrides. However, the retrieved material does not explicitly say whether the base model’s configured system prompt is appended, replaced, or inherited as a fallback

Important ambiguity in the documentation

The Chat Parameters page also has an “Optimize System Prompt Settings for Maximum Flexibility” section that says an account-level prompt can work “in conjunction with” a chat-level or model-level prompt. That wording is less precise than the hierarchy section’s override language; it does not explain the actual message assembly or concatenation behavior.

Bottom line: These are useful references for your Reddit post, but I still cannot verify base-to-workspace prompt inheritance from them. The precise question to attach is:

Questions:

1. When a workspace model has its own system prompt, does Open WebUI append the base prompt, replace it, or use it only if the workspace prompt is empty?
2. Where should shared research/citation instructions live: only on the base model, in every workspace model, or in another shared injection mechanism (eg. An Account Level System Prompt)?