Repository navigation
Releases: nmamizerov/assemblix
Releases · nmamizerov/assemblix
Release list
v0.2.46
0.2.46 (2026-10-29)
Features
- add Jev classification node (0472b4a)
- api: add a LiveAvatar LITE avatar adapter (b73abf1)
- api: add execution cost summary endpoint (12b741f)
- api: add execution cost summary endpoint (f36a215)
- api: add Jev node config schema (9486037)
- api: add Jev system node (9e98f39)
- api: add TypeSafe credential type for the Jev node (9323c99)
- api: add TypeSafe System One client (882d394)
- api: default credential listing to the API key's project (0445261)
- api: register LiveAvatar as a voice-agent avatar provider (5fd58ae)
- api: stream avatar audio to LiveAvatar over its session socket (62fb02e)
- bitHuman avatars for voice agents (ee6df99)
- LiveAvatar avatars for voice agents (974e024)
- web: add Jev node form and TypeSafe credential type (ab26098)
- web: pick LiveAvatar avatars for voice agents (d472dd2)
Bug Fixes
- api: always stop a LiveAvatar session before closing its socket (3df9dc0)
- api: validate Jev config at run time and keep structured state (7e74b29)
- give avatar joins 45 s and let the browser wait on the server's timeout (9a14b56)
- give avatar joins 45 s and let the browser wait on the server's timeout (d54d244)
Documentation
Refactoring
- api: let avatar vendors supply their own audio output and stop hook (8c64b5e)
v0.2.45
0.2.45 (2026-10-20)
Features
- api: re-ask Smart Turn with a decaying threshold and signal pauses (fee4a73)
- api: start cascade replies speculatively during pauses (5c44809)
- speculative replies and decaying Smart Turn for faster cascade turn-taking (ff3b87b)
- web: show speculative turn-taking details and add its toggle (01c5ff8)
Bug Fixes
- agent: close history ending with a model turn with a user turn (543ab51)
- agent: close history ending with a model turn with a user turn (a62af56)
- agent: raise chat history limit from 20 to 60 messages (50e00db)
- agent: raise chat history limit from 20 to 60 messages (118bd41)
- api: harden speculative cascade turn-taking (29b3f29)
- api: name the stalled stage when an avatar join times out (6f5c6ae)
- api: name the stalled stage when an avatar join times out (e5524e8)
v0.2.44
v0.2.43
v0.2.42
v0.2.41
0.2.41 (2026-10-03)
Features
- llm: refresh OpenAI and Gemini model catalogs to the October 2026 lineup (19c6d6f)
- llm: refresh OpenAI and Gemini model catalogs to the October 2026 lineup (edb246d)
- spoken-style default prompt and late barge-in for cascade voice agents (8adca3b)
- web: pre-fill voice agents with a spoken-style system prompt (30444a4)
- web: surface the cascade brain's reasoning controls (00b690e)
- web: tune the cascade brain's generation settings (f0a285b)
Bug Fixes
v0.2.40
v0.2.39
0.2.39 (2026-10-03)
Features
- api: add cascade mode to the voice agent config (53459b3)
- api: add cascade turn detector (VAD gating + Smart Turn) (7bbdccd)
- api: add Silero VAD and Smart Turn v3 models for cascade turn detection (de86c56)
- api: add the cascade conversation bridge (90c7ab4)
- api: add the prompt brain for cascade voice agents (7b8a59d)
- api: add Yandex SpeechKit v3 streaming recognition adapter (3afd065)
- api: cascade mode for voice agents (STT → turn detection → LLM → TTS) (b592460)
- api: make cascade voice agents callable (ff31c91)
- api: record per-stage turn timings for cascade voice calls (50270c6)
- api: share one Yandex TTS channel per call and add economical chunk mode (1988399)
Bug Fixes
- api: build the voice bridge before opening the session row (1c6e325)
- api: close the shared TTS channel when the inner bridge fails to connect (35db1e9)
- api: close the STT stream on a failed cascade connect and keep caller cancellation (20b9a1f)
- api: harden cascade bridge turn and interrupt handling (471edd1)
- api: report cascade turn timings once, with firstAudioMs (e4e1cf5)
- api: stop Yandex TTS synthesis on barge-in with a shared channel (6c635b5)
v0.2.38
0.2.38 (2026-09-25)
Features
- api: add LiveKit settings and SDK dependencies (e382d6a)
- api: let the voice runtime drive server-side playback channels (96bd3e3)
- api: let voice agents carry an avatar config (e892847)
- api: mint LiveKit participant tokens and delete rooms (84f11c9)
- api: open an avatar call's LiveKit room with agent, vendor and caller (bafa9ba)
- api: return a LiveKit room when minting an avatar call (b548f73)
- api: run avatar voice calls through LiveKit (f062ca7)
- api: start Anam avatars inside a LiveKit room (cbebf2b)
- api: stream agent audio to an avatar over LiveKit data streams (d0adbc6)
- deploy: add optional LiveKit tier for voice-agent avatars (7926a3a)
- lip-synced avatars for voice agents over LiveKit (e26fd47)
- web: configure an avatar on a voice agent (53d67ee)
- web: place avatar test calls over LiveKit (0c3c23f)
Bug Fixes
- api: bound avatar room teardown to five seconds (0b00393)
- api: bound the whole avatar join by the join timeout (27591f0)
- api: cancel both LiveKitChannel drain tasks before awaiting either (2c81b0a)
- api: cancel connect/prepare when run() itself is cancelled (766ed8f)
- api: give every failed avatar call a close reason (ac085ee)
- api: hand out LIVEKIT_PUBLIC_URL as a WebSocket URL (da89a46)
- api: let the caller publish only its microphone in an avatar room (bbb7979)
- api: never raise from LiveKit room deletion (ba1a2c3)
- api: require the avatar model on a voice agent's avatar (b9d91e2)
- api: tear the avatar down before the final hook runs (bc8ba0e)
- api: truncate a finished reply when the caller barges in during playback (f751f35)
- pass AVATAR_JOIN_TIMEOUT_SECONDS through the app compose file (aedb27f)
- web: let session.closed drive avatar teardown before a stale room can (e366c55)
- web: require the avatar model before saving a voice agent (94641f0)
Documentation
- document voice-agent avatar calls and the LiveKit tier (2c732cd)
- note the Gemini Live limit on interrupting an avatar (515ca8e)