Smarter Agents: Adaptive Thinking, Feedback, Model Control & Learning

A major upgrade to agent intelligence and user control. Agents now think deeper on complex tasks, learn from your feedback, and can be individually tuned with model preferences and cost limits.

• ›Adaptive thinking: agents automatically reason more deeply on complex tasks and respond faster on simple ones
• ›Per-agent model control: assign each agent a preferred model tier (e.g. use the fastest model for simple lookups, the most powerful for code generation)
• ›Cost guardrails: automatic per-request budget limits based on your plan tier prevent unexpected costs
• ›Feedback buttons: thumbs up/down on any response — your feedback improves agent performance and memory confidence over time
• ›Improved learning: agents now save successful tool patterns to a knowledge base even without explicit corrections, making them better at recurring tasks
• ›Chat agent creation parity: create agents via chat with team grouping and integration access — same fields as the UI form
• ›Progress updates: see what subagents are doing during long-running delegated tasks