LDCon / Agent operations
Agent settings
Choose computation profiles and how operational updates reach you.
Loading saved settings…
Difficulty routing
Profile IDs are arbitrary names. This table alone maps each card difficulty and agent interface to a profile. A card may use the default difficulty. Explicit per-run profile selection takes precedence; active tasks keep their admitted profile.
Computation
Execution profiles
Define profiles here, then assign them in the difficulty routing table above. Names such as basic or deep in a Profile ID have no special meaning.
Profiles select a provider, exact model and resource budgets. Saved changes apply to new sessions and tasks; active runs keep their selected profile.
Model and effort choices come from the locally verified capability list. Account access is checked when a task connects. Provider default leaves effort to that provider. Existing running tasks are unchanged.
Resource budgets: Max turns / Input bytes
Max turns caps the task’s model/decision rounds; it does not count hourly monitor wakeups. Input bytes caps the UTF-8 task instructions and evidence bundle, not tokens or total conversation history. Keep the saved budgets unless a task reaches its limit. The current adjustment profiles use 64 turns and 262144 bytes; reducing these can prevent a useful study. These budgets never expand device authority.
Communication
Notification policy
Automatic messages are brief English summaries. Human replies retain the original request language. Detailed evidence stays in local records at every preset.