Chat modes in Optimizely Mark chat

Every conversation with Mark (Optimizely Agent Platform) runs in a chat mode. Chat modes let you control how much context Mark chat pulls into each response. Choose a mode to balance speed, quality, and credit cost for each task. Mark chat processes every message you send in one of three chat modes: Auto, Fast, or Enriched. All three modes use your skills.

Switch between modes at any time using the chat mode selector in the chat input bar. The selector displays the current mode, such as Auto. Your selection applies to the next message you send.

Auto mode evaluates each prompt you send and determines which capabilities produce the best response. It does not treat Fast and Enriched as a single either-or choice. Instead, it decides separately whether to include knowledge retrieval, memory, and tools, and which model tier to use. For a straightforward question, Auto mode can skip retrieval and memory while still applying the appropriate skills. A complex content request receives the full set of capabilities.

Auto mode requires high confidence before it skips any enrichment capability. When uncertain, Auto mode includes more context rather than less. You get efficient responses for simple prompts without losing quality on complex ones.

This per-message optimization also reduces credit consumption. Mark spends credits only on the capabilities each prompt needs, rather than applying full-context processing to every message. For most teams, Auto mode balances response quality, speed, and credit efficiency without manual switching.

Auto mode is the default. Optimizely recommends it for most users. See How Auto mode routes prompts.

Fast mode generates responses using your skills and lightweight application context only. It skips knowledge retrieval (Retrieval-Augmented Generation, or RAG) and memory. Mark does not pull in brand guidelines, campaign context, or other stored information. Response quality remains high, but without the added context that Enriched mode provides. Fast mode also uses fewer credits per response. Because Mark skips these lookups and works from a shorter prompt, Fast mode returns responses faster.

Use Fast mode in the following situations:

Enriched mode generates responses using the full set of available context, including your skills, knowledge retrieval (RAG), and memory. Mark considers your brand guidelines, campaign context, and other relevant information when forming a response. Because Enriched mode retrieves and processes additional context, it uses more credits per response than Fast mode.

Use Enriched mode in the following situations:

Chat mode affects credit usage, which counts toward your organization's Mark consumption. Auto mode optimizes this consumption automatically. The following table compares the three chat modes:

Change the chat mode at any time. The selection applies to your next message.

The selector displays the mode you selected.

Before Auto mode generates a response, it runs a lightweight classification step. The step does the following:

The routing decision happens in milliseconds and does not add noticeable latency to your response. Auto mode is deliberately cautious about skipping context. It needs strong confidence to skip a capability, but only moderate confidence to include one. This bias favors quality over cost savings.

For a deeper look at the design thinking behind Auto mode and prompt routing, see Not every question deserves a PhD.