Profiles & Orchestration

A profile is a saved assistant: how it works through requests, its models, tools, memory, and the computers it runs on. General Assistant decides everything for you. Programming uses the Team strategy. Research focuses on looking things up. In advanced mode, Profiles & Orchestration in the sidebar lets you create, rename, duplicate, and edit them, and its editor opens across the page with the profile's name and four tabs: Models, Tools, Knowledge and memory, and Strategy and computers. Models shows the roles in use, with Show all roles for the rest; Tools has a switch for each kind of tool, with Per-tool permissions below. A tool a profile doesn't list stays off. Built-in profiles cannot be deleted; More actions → Reset to defaults puts one back to how Toskar ships it.

Strategy is how the profile works: Auto (quick questions get one model, bigger requests are planned and checked), Single model, Planner + workers (requests with several parts are always worked through in parts), or Team (a planner splits each request, workers do the parts, and a reviewer checks the answer; quick questions are still answered directly).

Models assigns a model, and optionally a computer, to each role: Primary writes the answer, Fast and Coding answer quick and coding requests when the chat is on Auto, and Planner, Worker, and Reviewer do their part of a plan. Automatic uses the model chosen in the chat. Fallback order lists the models to try, in order, when the answering model fails.

Capabilities turn tool groups on and off, and each tool has its own permission. A profile can list knowledge sources that every chat with it searches. Computer selection can be Automatic or Prefer this computer; Only this computer keeps every turn here, and each paired computer can be Preferred or Never use.

Orchestration tunes how the profile works: Reasoning level (its effort when a chat leaves effort on Auto), Planning (when a request has parts, always, or never), Workers and Parallelism (how many parts, and side by side or one at a time), Verification (check and report, or check and correct once or twice), Tool calls, Memory (off for this profile), Context budget (how much of the model's window earlier messages may use), Fallback (show the failure instead of answering on another model), Retries (how many times a failed turn is tried again, first on another computer with the same model), and Time limit. Empty fields keep the defaults.