Cortex is one engine rather than several separate assistants. The header chat, the admin console and generation inside modules such as Learning all run on the same Claude-based engine (Anthropic, called directly). Two controls shape each request: effort and budget.
Effort
- Low: fast answers for lookups, rewrites and short summaries.
- Medium: the sensible default for most analysis and drafting.
- High: multi-step reasoning across many records, at higher cost and latency.
Budget
Budget is the ceiling on what a request may consume. Together with effort it determines how far Cortex is allowed to go before returning an answer. Agent playbooks add their own spendCapCredits so an autonomous run cannot exceed a limit you chose.
Choosing settings
- 1Start at the default effort for your task.
- 2If the answer is shallow or misses relationships between records, raise effort one level and re-run.
- 3If simple questions feel slow or costly, lower effort.
- 4For agents, set the model to auto unless you have a reason to force sonnet or opus.
Usage draws on your workspace credit balance. Administrators can review consumption and set budgets from the credit and billing settings.