Studio
Pick how the public side runs, then press Start. Everything unloads after 5 minutes without public activity.
Operational mode
Single agent
Combo agent
Describe what each thinker is good at: the manager reads these descriptions to choose. Thinkers load only when assigned a prompt, unload at once when the next prompt goes to a different one, and after 1 idle minute.
Parallel agent
At most this many chats generate at the same time; the rest wait in the priority queue. Ollama serves them from one loaded copy, so also make sure OLLAMA_NUM_PARALLEL (set by setup.sh, default 4) is at least this number. Each slot costs extra RAM for its context.
Context window
Sets Ollama's num_ctx for every model of the chosen mode, from 1k (1,024 tokens) to 512k (524,288 tokens). Ollama reserves the whole window in memory when a model loads, so a bigger window costs more RAM: press Test Setup to see how much. A change applies at the next Start.
Users
Accounts for the public side. Level 1 Normal: 30,000 tokens/hour. Level 2 Advanced: 60,000 tokens/hour, queued ahead of Normal. Level 3 Admin: unlimited, first in the queue, can manage this studio.
| User | Level | Tokens this hour | Window resets | Lifetime tokens | Last seen | Status | Actions |
|---|---|---|---|---|---|---|---|