Guides
Creating Custom Prompt Libraries in Open WebUI
Save, organise, and share reusable prompt templates in Open WebUI so your team gets consistent AI responses every time.
Creating Custom Prompt Libraries in Open WebUI
Prompt libraries turn ad-hoc AI usage into repeatable, reliable workflows. Instead of typing the same instructions from memory — and getting slightly different results each time — you store well-tested prompts in Open WebUI's template system and reuse them with one click.
This guide covers how to design, organise, and share prompt libraries for yourself and your team.
Why Prompt Libraries Matter
Without a library, every interaction depends on your ability to write a good prompt on demand. That works for simple questions, but complex workflows — document analysis, code review, content drafting — benefit from carefully built prompts tested over multiple sessions.
A prompt library gives you:
- **Consistency** — the same prompt produces the same quality
- **Speed** — no retyping or remembering complex instructions
- **Sharing** — team members benefit from your best prompts
- **Iteration** — refine prompts over time and save improvements
If you are new to prompt design, read Prompt Tuning for Local LLMs Without Overcomplicating Things first.
Creating Your First Prompt Template
Step 1: Navigate to the prompt library
In Open WebUI, open the prompt panel from the sidebar.
Step 2: Design the template
A good prompt template has four parts:
**Role:** You are a senior technical writer.
**Task:** Review the following document and produce a structured summary.
**Format:**
- Key findings (3-5 bullet points)
- Important details (2-3 paragraphs)
- Action items (numbered list)
**Constraints:** Be concise. Use UK English. Do not add information not present in the source.
Every template should include a clear role, the task, the output format, and constraints.
Step 3: Add variables
Open WebUI supports template variables so one prompt works for different inputs:
**Task:** Translate the following text from {{source_language}} to {{target_language}}:
{{text_to_translate}}
**Requirements:** Maintain the original tone.
Organising Your Prompt Library
Use naming conventions so prompts are easy to find:
| Prefix | Example | Purpose |
|--------|---------|---------|
| summarise_ | summarise_meeting_notes | Condense information |
| analyse_ | analyse_code_security | Examine for issues |
| draft_ | draft_blog_post | Generate new content |
| translate_ | translate_to_spanish | Convert between languages |
| review_ | review_document_clarity | Critique and suggest improvements |
Categorise by workspace. Engineering prompts belong in the Engineering workspace. For workspace management, see Open WebUI Workspaces for Team Document Collaboration.
Advanced Prompt Techniques
Chain prompts together
Complex tasks benefit from multi-step prompting:
1. analyse_meeting_transcript — extracts key points
2. draft_action_items — converts decisions into tasks
3. format_email_summary — wraps results in a template
System prompt integration
Open WebUI lets you set system-level prompts that apply before templates.
Sharing Prompts Across a Team
Prompts created in a workspace are visible to all workspace members. Encourage team members to comment on prompts, version similar ones, and archive unused ones.
Maintaining Your Prompt Library
Review prompts every month. Remove unused ones, update outdated ones, and test prompts when you change the underlying model.
Troubleshooting
If the model ignores format instructions, use models known for instruction following such as Phi-4 or Llama 3.2. If variables do not render, check that names use double curly braces.
Conclusion
A well-organised prompt library is the difference between hoping for good AI output and reliably getting it. Start with five templates for your most common workflows, refine them over a week, and expand from there.
FAQ
Can I export my prompt library?
Open WebUI stores prompts in its database. Regular database backups preserve your prompt library.
Do prompts work across different models?
Most prompts transfer within the same model family. Switching families may require adjustments.
Can prompts include example outputs?
Yes. Including a few-shot example helps models understand the expected format.
**Sources:**
