How to ignore prompts generated by open-webui internals in pipelines

[JanciJakGenerAIMage]

I am struggling with basic implementation. I have a basic scaffold running, but there are multiple messages coming from open-webui internals that mess up my business logic. Is there any way how to ignore them, or force them not to be in my pipeline?

For example:

'### Task:\nBased on the chat history, determine whether a search is necessary, and if so, generate a 1-3 broad search queries to retrieve comprehensive and updated information. If no search is required, return an empty list.\n\n### Guidelines:\n- Respond exclusively with a JSON object.\n- If a search query is needed, return an object like: { "queries": ["query1", "query2"] } where each query is distinct and concise.\n- If no search query is necessary, output should be: { "queries": [] }\n- Default to suggesting a search query to ensure accurate and updated information, unless it is definitively clear no search is required.\n- Be concise, focusing strictly on composing search queries with no additional commentary or text.\n- When in doubt, prefer to suggest a search for comprehensiveness.\n- Today\'s date is: 2024-11-22\n\n### Output:\nJSON format: {\n  "queries": ["query1", "query2"]\n}\n\n### Chat History:\n<chat_history>\nUSER: Can you tell me a joke?\n</chat_history>\n

Title prompt:

'Create a concise, 3-5 word title with an emoji as a title for the chat history, in the given language. Suitable Emojis for the summary can be used to enhance understanding but avoid quotation marks or special formatting. RESPOND ONLY WITH THE TITLE TEXT.\n\nExamples of titles:\n📉 Stock Market Trends\n🍪 Perfect Chocolate Chip Recipe\nEvolution of Music Streaming\nRemote Work Productivity Tips\nArtificial Intelligence in Healthcare\n🎮 Video Game Development Insights\n\n<chat_history>\nUSER: Can you tell me a joke?\nASSISTANT: test_pipeline response to: Can you tell me a joke?\n</chat_history>

Tag prompt:

### Task:\nGenerate 1-3 broad tags categorizing the main themes of the chat history, along with 1-3 more specific subtopic tags.\n\n### Guidelines:\n- Start with high-level domains (e.g. Science, Technology, Philosophy, Arts, Politics, Business, Health, Sports, Entertainment, Education)\n- Consider including relevant subfields/subdomains if they are strongly represented throughout the conversation\n- If content is too short (less than 3 messages) or too diverse, use only ["General"]\n- Use the chat\'s primary language; default to English if multilingual\n- Prioritize accuracy over specificity\n\n### Output:\nJSON format: { "tags": ["tag1", "tag2", "tag3"] }\n\n### Chat History:\n<chat_history>\nUSER: Can you tell me a joke?\nASSISTANT: test_pipeline response to: Can you tell me a joke?\n</chat_history>

These messages appear multiple times.

There is only one desired message, that I wanted to use and that was the last user prompt:
[{'role': 'user', 'content': 'Can you tell me a joke?'}]

[gesaleh]

I'm facing the same issue

[benx13]

same

[baerenfreund]

@JanciJakGenerAIMage @gesaleh @benx13

I encountered the same issue and found an easy solution:

Set the Task Model Environment Variable:

• Define TASK_MODEL_EXTERNAL (and optionally TASK_MODEL) with your preferred model instead of using the pipeline directly for task generation.
• Alternatively, configure these in the admin settings, or even disable Tasks if needed in the admin settings.

https://docs.openwebui.com/getting-started/advanced-topics/env-configuration/#tasks

Notice that we have to set the Pipeline URL as an OpenAI API connection. This configuration causes the system to interpret pipeline calls as OpenAI API requests, defaulting to TASK_MODEL_EXTERNAL. If this variable isn’t set (which is the default), the pipeline itself will be used for task generation.