Settings · PeerUp AI Playground

Provider & model

Mirrors prod's A/B model + provider buckets. Empty env vars hide the option's functionality.

LLM provider

Temperature

Gemini model

Bedrock model

STT / TTS

STT provider

TTS provider

Note: when LLM = Gemini AND STT = Gemini audio, the playground sends one Gemini call with audio inline (the prod bucket-A pipeline). Other combos transcribe first and then call the LLM with text, like the prod bucket-B / fallback pipelines.

First-message variant (intro turn)

Matches prod's FIRST_EVER_AI_REPLY_EXPERIMENT (peerup-backend/services/ChatWithAI/controller/get-ai-replies.js).

Memory

Prod has two memory flows: (a) inject the user's memory into every prompt turn, (b) auto-extract memory after each call via a separate Gemini pass and merge it into the user's stored memory. Toggle each independently here.

Inject the user's stored memory into the system messages Auto-extract & save memory after each call (Gemini-based, like prod's updateMemoryContextForAIChat)