DeepSeekTool Switching in Composer
DeepSeek composer: Instant/Expert tier toggle, DeepThink and Search chips in the card, and attach with tier-specific limits.
Instant default with Search on, DeepThink available, and attach plus send in the card.
What's happening
DeepSeek uses a centered composer card with Instant/Expert tiering above the bar and DeepThink/Search chips inside the card. Search defaults on in Instant, attach limits vary by tier, and Expert simplifies the capability set.
Patterns
Speed vs quality tiers with header copy that tracks selection
Reasoning and search as armable chips with outcome tooltips
Tier-specific attach limits and parsing state inside the card
UX Insights
- •Search ships on by default in Instant for research-ready first sends
- •DeepThink tooltip explains reasoning in plain language, not model jargon
- •Expert mode removes Search from the card when users switch tiers
- •Instant attach is text-extraction only with explicit count and size caps
- •File chips show parsing before send without leaving the compose surface
Design Decisions
DeepSeek optimizes for honest tiering on one card: fast searchable Instant by default, Expert plus DeepThink for depth, with attach rules that protect the cheap tier.
More case studies
ManusTask composer & modes
Manus assignment bar: connector suggestions, + attach menu, skills, cloud computer, agent tiers, voice input, and mode-specific flows like Slides.
GLM (Z.ai)Tool Switching in Composer
GLM (Z.ai) composer: Magic Design, Full-Stack, and Write Code pills; Search vs Advanced Search; GLM-4.7 through 5.2; Deep Think Max; separate Agent home with Discord/Lark/WeChat.
KimiTool Switching in Composer
Kimi composer: agent pills that rewrite the bar into Websites, Docs, Sheets, Design, and Swarm; Instant/K3/K3 Swarm; + for files plugins skills and search; slash invoke.
ChatGPTTool Switching in Composer
ChatGPT allows users to select and activate different AI capabilities like Code Interpreter, DALL·E, and Web Search directly from the composer interface.