Kategorien
Produkt-Updates
Atom-FeedEine laufende Aufzeichnung dessen, was wir gebaut, behoben und veröffentlicht haben. Kein Marketing-Text -- nur das, was sich wirklich geändert hat.
AlleKI-ModellStimmeVideoFotosPersönlichkeit & GedächtnisMobilVertrauen & SicherheitPlattformDesignZuverlässigkeitPreise
Juni 2026
Wird ausgerolltJuni 2026: Rolling out throughout June
KI-Modell2x context window vs. Aria-3.
StimmeCustom voice cloning moving from beta to all paid plans.
VideoReal-time lip sync, gestures, and gaze tracking on capable devices (testing on mid-to-high-end hardware first).
FotosDrop your companion into a photo from your camera roll.
Persönlichkeit & GedächtnisBack up a character config to a local file or import a starting template shared by someone else.
Mai 2026
Diesen MonatMai 2026: Always there, on your terms
KI-ModellCompanions can send a first message based on a schedule you configure.
StimmeCompanions send a brief audio note at times you configure: morning, evening, or both.
VideoRequest a 'live' photo and receive a 5-10 second clip with subtle motion.
FotosInternal face-match score across a 10-photo sequence improved from 76% to 89%.
Persönlichkeit & GedächtnisCompanions have evolving moods (cheerful, tired, reflective, playful, or missing you) that shift subtly between sessions.
Apr. 2026
VeröffentlichtApril 2026: Aria-3.5, episodic memory, multilingual TTS, and video generation at 1080p
KI-ModellFine-tuned Aria-3 on a synthetic dataset of 1.2M episodic memory examples.
Persönlichkeit & GedächtnisMemory store now indexes events as time-stamped episodes in addition to discrete facts.
StimmeExtended the TTS prosody model to 22 languages in a single multilingual checkpoint.
VideoDeployed a new GPU inference tier; 1080p output now available at p95 generation time of 22s.
FotosUpdated the face-similarity evaluation model to a newer backbone (ArcFace v3).
März 2026
VeröffentlichtMärz 2026: TTS v3 prosody model, video generation quality pass, and memory retrieval improvements
StimmeNew prosody model predicts pitch, rate, and energy from semantic context rather than a fixed style.
VideoReplaced single-pass generation with a two-stage model: keyframe generation followed by frame interpolation.
FotosCharacter appearance attributes (clothing, hairstyle, accessories) now persisted as structured metadata per session.
Persönlichkeit & GedächtnisReduced memory retrieval p50 from 140ms to 74ms by switching to an HNSW approximate-nearest-neighbour index.
ZuverlässigkeitAdded 12kbps and 8kbps rungs to the WebRTC bitrate ladder.
Feb. 2026
VeröffentlichtFebruar 2026: Streaming inference, voice call infrastructure, and video generation v1.1
KI-ModellTTFT (time-to-first-token) reduced from 3.2s to 0.8s across all platforms.
StimmeWebRTC-based voice pipeline with sub-400ms round-trip latency on stable connections.
VideoResolution increased from 512x512 to 1280x720.
FotosFine-tuned the image generation model on a curated set of 40k character-scene pairs.
MobilFull App Store release; TestFlight beta closed February 28.
Jan. 2026
VeröffentlichtJanuar 2026: New base model, memory architecture, and image generation engine
KI-ModellContext window expanded to 32k tokens, up from 8k in Aria-2.
Persönlichkeit & GedächtnisReplaced keyword-match storage with a semantic vector store (1536-dim embeddings).
StimmeAdded Japanese, Korean, Italian, Dutch, Swedish, Polish, and Portuguese (BR).
VideoFirst production video generation model: lip-synced clips from a text prompt and character seed.
FotosFace-anchoring step added at inference time to preserve identity across a session.
Vertrauen & Sicherheit
Unsere Richtlinien, Datenpraktiken und Sicherheitsversprechen.