Open WebUI
🧩 1. “Functions” (what they actually are)
Think of Functions = tools the model can call
Not functions in a coding sense (well… under the hood yes), but practically:
“when the model needs to do something external”
Examples:
- web search
- calculator
- call an API
- run a script
So instead of just:
LLM = text in → text out
you get:
LLM = can take actions
🧠in your setup
You’ll eventually use functions for things like:
- “save this to Obsidian”
- “trigger n8n workflow”
- “search my knowledge base”
- “fetch latest news”
👉 don’t configure anything here yet Just know:
this is your future automation layer
🧪 2. “Evaluations”
This one sounds fancy but is simpler:
Evaluations = testing / benchmarking prompts & models
Used for:
- comparing models (“which is better at summarising?”)
- testing prompt quality
- regression testing
🧠for you right now
👉 ignore it
You’ll care about this later if you:
- optimise prompts
- compare Mammouth models
- build repeatable workflows
3. web search
- DDGS → free, scrappy, good enough ✅
- Brave / Tavily → paid, stable, cleaner 💰
- SearXNG → self-hosted, flexible 🔧
âś… recommendation for you (right now)
Stick with DDGS and move on with the tutorial.
You’ve already got enough moving parts:
- Open WebUI
- Mammouth API
- future Obsidian integration
4. Backup
need to setup a backup pipeline for this and other self hosted tools
Mammouth
Web UI and Mobile Clients
CLI tool
Obsidian
Livesync plugin
ChatGPT MD plugin
API Temperature
| Temperature | Behaviour | |
|---|---|---|
| 0.0 | deterministic, same answer every time | |
| 0.2–0.4 | precise, factual, safe | for coding / config (like Docker, APIs) |
| 0.5–0.7 | balanced (default sweet spot) | good general use |
| 0.8–1.2 | creative, varied | for writing / brainstorming |
| 1.5+ | chaotic, sometimes nonsense |
can plugin support different temperature within template / YAML frontmatter?
Tool Compatibility
what TOOLS does it support for what models?
Here are the main tools chatgpt 5.4 nano access to in this chat: - Web - search_web – search the public web for information. - fetch_url – fetch and extract the main text from a specific webpage URL. -
Files / knowledge (internal) - Knowledge base tools: list_knowledge_bases, search_knowledge_bases, query_knowledge_bases, search_knowledge_files, query_knowledge_files, grep_knowledge_files, view_knowledge_file - Notes tools: search_notes, view_note, write_note, replace_note_content - Chat history tools: search_chats, view_chat -
User memory (persistent) - list_memories, search_memories, list_memory_paths, read_memory_path - add_memory, update_memory (batch), replace_memory_content, delete_memory -
Automation / scheduling - create_automation, update_automation, list_automations, toggle_automation, delete_automation -
Calendar - search_calendar_events, create_calendar_event, update_calendar_event, delete_calendar_event -
Code execution - execute_code – run Python in a sandbox (no installing packages) If you want, I can use web search + URL fetching to find and summarise recent UK political news—tell me the topic (and whether you mean Nigel Farage specifically).