Chat API
How to useThe main model behind TA's replies — the one that lets TA talk.
How to setAPI Setup → "Chat", fill in the endpoint, key, and model name.
How much?Start with maxTokens 4096, temperature 0.7; lower if it lags, or switch to a smarter model.
Embedding API
How to useTurns your memories / knowledge (the knowledge folder, incl. shared memories) into vectors. TA recalls via hybrid retrieval (keywords + vector semantics) — once configured, TA can recall related old memories by meaning even if you didn't use the exact words.
How to setAPI Setup → configure a separate Embedding entry (TA won't reuse your Chat one even if it supports embeddings).
How much?Use your provider's recommended embedding model; the default is usually enough. If missing, TA falls back to keyword-only recall.
Web Search API
How to useLets TA search the live web for current info, instead of relying on memory alone.
How to setAPI Setup → "Web search", fill in endpoint and key.
How much?Provider defaults are fine; keep an eye on quota if you search a lot.
TA Migration
How to usePackage a TA with its chats, memories, and settings into a file, then import it with one tap on another phone.
How to setApp Settings → TA Migration → Export zip; choose Import on the new device.
How much?One TA per zip; circles and memory travel with it. Export before switching phones or cleaning up.
Export Logs
How to useExport runtime logs to troubleshoot issues or report feedback.
How to setApp Settings → Export Logs.
How much?Logs include your API key (encoded but reversible) — a warning appears before export; share with care.
Background Keep-Alive
How to useLets TA keep working even when the app is closed (sending proactive messages, organizing memories).
How to setSettings → Keep-Alive; enable the persistent notification and follow the prompts to disable battery optimization and allow auto-start.
How much?The more aggressive your phone's background kill, the more you need it. If TA stops messaging you, check this first.
Nickname
How to useWhat you call this TA — visible only to you.
How to setFill it in when creating the TA.
How much?Whatever you like.
Name
How to useTA's display name; written into the persona file (persona.md).
How to setFill it in when creating the TA.
How much?Decide before finishing — after creation it can only be changed in the .md file.
Custom Conversation ID (UUID)
How to useGives this conversation a unique ID for migration / binding; different TAs use different IDs so data never mixes.
How to setFill it in when creating the TA (usually left as is).
How much?Not really needed — generally ignore it.
Pause Button
How to useA "Pause / Resume" button beside the input box (long-press for a hint). Tap it to make the AI stop, so you can send your message in segments — one part, pause, then continue with the next.
How to setNo toggle needed — just tap the button.
How much?Use it when you want the AI to reply one message at a time.
Long-Press Undo Send
How to useLong-press a message you just sent and "Undo" appears in the menu.
How to setNo toggle — it appears automatically when the condition is met.
How much?Only works when the AI is paused and it's your last unanswered user message.
Segmented Sending Message Settings
How to useSplits TA's long replies into several messages by sentence / segment, so it reads like a real person texting.
How to setMessage Settings → Segmentation toggle; you can edit the segmentation rules.
How much?On for a more natural feel; off if it feels too choppy or you prefer one long message.
Memory · Master Switch
How to useWhether the whole memory system is active. Turn off to stop recording entirely, for pure chat.
How to setMemory System → master switch.
How much?On by default; turn off if you don't want to be remembered.
Memory · Context Trimming
How to useAuto-discards the oldest messages as the conversation nears its limit, making room for new topics.
How to setMemory System → Context Trimming; adjust the threshold.
How much?Automatic by default; a banner warns around 80% full. On keeps very long chats from being hard-truncated.
Memory · Daily Flush
How to useHow often new chats are batched into a same-day summary.
How to setMemory System → Daily Flush; set the interval.
How much?Defaults to a 2-minute batch; set 0 to disable. Bigger interval = cheaper & tidier; smaller = more timely but more fragmented and token-hungry.
Memory · DeepDream Distillation
How to useDistills fragmented memories into refined long-term memory, archived by core facts / key moments / recent status, auto-pruned when too large.
How to setMemory System → DeepDream; toggle and edit the prompt.
How much?On by default; keeps TA remembering you more accurately and longer.
Memory · Diary
How to useGenerates a daily diary entry on your real time each day.
How to setMemory System → Diary toggle.
How much?On by default; turn off if you don't want them kept.
Memory · Retrieval Mode
How to useDecides how TA "remembers" — via vectors (embeddings) or keywords.
How to setMemory System → Retrieval Mode.
How much?Vectors suit associative recall, keywords suit exact retrospection; once vectors are set up, hybrid is usually best.