Reference

clientConfiguration fields (per-session agent config)

The server sends clientConfiguration once per connect. The SDK keeps it and exposes it read-only as session.clientConfig. These are the fields the SDK reads:

Field	Example	What the SDK does with it
`isTapToTalk`	`false`	Push-to-talk vs open-mic. A fixed, per-agent choice, not a live toggle. Read as `capabilities.tapToTalk`. `startTapToTalk()` and `endTapToTalk()` throw `capability_disabled` when it is `false` (see the `tapToTalkStart`/`tapToTalkEnd` row in the events catalog for why mixing modes is unsafe).
`interruptionsEnabled`	`true`	Read as `capabilities.interruptions`. `false` means the user can't talk over the avatar.
`isWebSearchEnabled`	`false`	Read as `capabilities.webSearch`. Web-search tools produce `tool`/`tool_response` deltas only when it is `true`. When `false`, the agent can still say it will "look that up", but no `tool` segment fires.
`isScreenShareEnabled`	`false`	Read as `capabilities.screenShare`. `startScreenShare()` and `sendScreenShot()` throw `capability_disabled` when it is `false`. It becomes `true` when the agent's `screenShare` setting is on.
`isCameraAnalysisEnabled`	`false`	Read as `capabilities.cameraAnalysis`.

The payload also carries languageCode, showTranscription, audioMode, phoneMode, pauseConversationEnabled, shouldAggregateCurrentTurn, initialHtml, youtubeUrl, visualPhotos, visualVideos, agentPersonaName, userName, configuration and nluFeatures. The SDK passes them through on session.clientConfig and does not act on them.

Structured experiences (`force_experience` + `unisphere-tool`)

Scope: the structured-experience behavior below is what you get on the HTTP /assistant/converse path (headless/text integrations). The live avatar socket does not produce flashcards or summarization. See Two delivery paths.

force_experience on converse (e.g. "flashcards") is a hint, not a guarantee. The agent decides which structured widget(s) to emit based on the prompt and its intellect (the agent's configuration: prompts, tools and capabilities). Each comes back as unisphere-tool segments. The first carries metadata:{ widgetName, runtimeName }, then the content streams as a YAML-ish block (e.g. title: / questions:). For example:

force_experience:"flashcards" + a teachable prompt ("Teach me about video codecs") → both flashcards-tool and followups-tool runtimes in one turn (widgetName:"unisphere.widget.genie").
The same force_experience + a vague prompt ("show me something interesting") → only followups-tool.
So: render whatever runtimeName arrives; don't assume force_experience maps 1:1 to a widget. capabilities.generate_followup_questions:"on" independently yields the followups-tool.

Doc	Covers
Wire Protocol · Events Catalog	The `join`/`clientConfiguration` events these fields ride on
Wire Protocol	Back to the index