Organization administrators manage a single tenant: members, simulation configuration, tasks and scenarios (where the product exposes them), and org-specific policy. Unless the product limits scope further (for example by cohort), changes apply to every learner in that organization.

​
Where to find settings documentation

Use these pages as the authoritative reference for configurable fields:
DocumentUse it for
Simulation settings — full catalogTop-level fields on SimulationSettings / UpdateSimulationSettingsRequest: meaning, defaults, learner impact, and operational notes.
Simulation policy pathsNested simulation_policy.* settings (realism, routing, privacy, orchestration, memory, tools, and related areas).
Parameter documentation standardA repeatable template for formal reviews, pilots, or customer appendices when you need more than the catalog tables.
Implementation references: SimulationSettings and UpdateSimulationSettingsRequest in src/common/models.py; admin field mapping in src/apis/simulation_admin/settings.py.

​
Practices for safe changes

  1. Use a staging org first — Exercise scenario, chatter, and policy changes outside production.
  2. Watch cost and load — Chatter limits, compression, and context windows directly affect LLM usage and background work.
  3. Align with learners — Time-boxed modes, completion messaging, and approval flows should be clear before a cohort starts.
  4. Control Expert-level fields — Restrict who may edit Expert-only parameters; schedule changes when they can be reviewed and audited.

​
AI usage and pricing visibility

Organization admins can view tenant-scoped AI usage and set pricing used for estimated cost views.
  • Usage views include totals, trends, per-user breakdown, and request/model/feature splits.
  • Pricing controls include:
    • LLM input price per 1M tokens
    • LLM output price per 1M tokens
    • Embedding price per 1M tokens
    • Optional mem0 unit price
If no org-specific pricing exists, default baseline pricing is used. See AI usage and cost tracking.

​
Cost-control knobs in simulation settings

In addition to pricing rates, two org-level settings materially change monthly token/cost burn:
  • simulation_policy.ambient.chatter_hourly_token_budget (ambient chatter token guardrail)
  • psychometric_recalc_interval_messages (how often psychometric LLM scoring is recomputed)
Use these with caution:
  • lowering them reduces spend but can reduce background realism/granularity
  • raising them increases realism density but can materially increase cost

​
Quick Help voice assistant controls

In Workspace Configuration → Expert mode → Quick Help Voice, organization admins can tune LiveKit call assistance behavior:
  • quick_help_call_enabled — master toggle for learner access to Quick Help calls
  • quick_help_call_voice — default OpenAI voice used by the LiveKit worker
  • quick_help_call_persona — organization-specific coaching persona instructions
These values are org-scoped and apply to newly created Quick Help call sessions.