New Features
- Add a Report a problem option to the Studio profile menu and error page, and to the screen shown after a conversation ends. Reports reach the Beyond Presence team with your account details and an optional screenshot
- Add support for the ElevenLabs Eleven v4 and Eleven v4 Turbo voice models in agents
- Agents: Add GPT 6 Luna (
gpt-6-luna) as an OpenAI LLM option. The default stays GPT 5.4 mini - Add a signed session manifest to every cloud avatar session. The avatar sends it on the LiveKit data topic
bey-provenanceto mark its video as AI-generated, and you can verify it with our published public keys (Session Manifest) - Agents: Add 11 Managed Agent languages: Croatian, Georgian, Gujarati, Hebrew, Kannada, Marathi, Punjabi, Tagalog, Tamil, Telugu and Thai, for 48 in total (Languages)
- Agents: Agents using Deepgram speech recognition in Spanish, French, German, Hindi, Russian, Portuguese, Japanese, Italian or Dutch now use Deepgram Flux Multilingual, for faster turn-taking
- API: Add sending context to an ongoing conversation with
POST /v1/conversations/{id}/context, so your backend can tell the agent what the user is doing and have it respond right away (Growth Plan and above, Just-in-Time Context)
Updates
- Agents: Gemini Live agents now run on Gemini 3.8 Live
- Agents: Agents that speak with an ElevenLabs voice now use Eleven v4 Turbo by default, for more natural speech at the same low latency
- Studio: Training videos for new custom avatars are now rejected at upload if they are recorded below 720p or show more than one person (Custom avatars, Troubleshooting)
- Agents: Agents without an LLM configured now use Gemma 4 for lower latency, with OpenAI as a fallback. English-language agents now transcribe speech with Deepgram Flux, also for lower latency
- Agents: Agents now have better interruption handling, so short acknowledgements like “uh-huh” no longer cut the avatar off mid-reply
- Agents: OpenAI retires GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini on January 14, 2027. Agents specifying one of these models now fall back to GPT 5.4 mini instead
Bug Fixes
- Fix sessions not being billed for the time after the avatar last spoke. Note: Idle sessions now correctly count toward your usage too
- Agents: Fix agents sent to an external meeting showing an empty camera when the meeting bot joined before anyone else (External Meetings)
- Agents: Fix conversations using the first microphone in the list instead of your browser’s default microphone when you haven’t picked one
Breaking Changes
- API:
404 Not Foundresponses no longer includeavatar_idoragent_idnext todetail. The body now only containsdetail. If your client reads those keys, use the ID from your request instead (API Errors) - API: All plan-gated features now return
402 Payment Requiredinstead of403 Forbidden. This covers the usage limit, avatar slots, the Call API, region pinning, message tracking, and custom branding.403now only means the resource is not accessible with your API key, and429only means the concurrency limit. If your client branches on403for plan restrictions, switch to402(API Errors)
New Features
- Add sending agents from Studio to ongoing Zoom, Google Meet, Microsoft Teams, and Webex meetings, with an optional Group Call mode that uses wakeup keywords (beta, paid plans, External Meetings)
- Add support for bringing your own ElevenLabs agent: give an existing ElevenLabs Conversational AI agent a Beyond Presence avatar via the API (ElevenLabs Agent)
- Add ElevenLabs agent setup in Studio: connect and manage ElevenLabs accounts, select ElevenLabs Agent as the engine, and paste an agent ID or link (ElevenLabs Agent)
- Add a ready-to-run ElevenLabs agent example to the bey-examples repository (ElevenLabs Agent)
- Add support for speech-to-video sessions of up to 24 hours (previously 3 hours). Sessions still end after 1 hour without speech (Concurrency & Quotas)
- Add usage endpoints to the API:
GET /v1/usagereturns your organization’s usage in the current billing period, andGET /v1/usage/sessionslists the usage of each speech-to-video session in a date range. Both require an API key of your default workspace
Updates
- Avatars: Faster session startup and quicker first greeting in general. Custom avatars with a cloned voice also start faster, since the voice clone is now reused across conversations
- Avatars: Speech-to-video sessions keep running during server maintenance instead of being interrupted
- Agents: Wakeup Mode now wakes the agent only when someone actually speaks to it, by its name or one of its triggers. The agent always answers to its name, so an empty
triggerslist is now valid (Wakeup Mode) - Agents: ElevenLabs agents now work with every ElevenLabs audio output format, including
ulaw_8000(mu-law, used for phone calls). Your ElevenLabs agent no longer needs to output PCM audio (ElevenLabs Agent) - Agents: Managed Agent conversations start faster. The agent now joins together with the avatar when the room is created, so the first greeting arrives about 1.5 seconds sooner
- Agents: The avatar now leaves within a few seconds after the caller hangs up, instead of staying in the room for about 25 seconds
- Agents: The call screen now shows when a conversation ended because it hit the time limit or idle timeout, instead of a generic failure
- Agents: Direct conversation links now enforce your plan’s usage and concurrency limits, and show a clear message when all avatars are busy or the agent can’t take more calls (Concurrency & Quotas)
- Agents: With message tracking turned off, conversations no longer send any data to third-party analytics or error monitoring services
- Studio: Studio now accepts training videos up to 1GB when you create a video avatar (previously 500MB) (Custom avatars)
- Studio: Group Call no longer requires keywords. The agent answers to its name, and keywords add extra names (Group calls)
- API: Deleting an avatar, external API configuration or knowledge file that an agent still uses is now refused instead of breaking the agent (API Errors)
- API: Requests that reference a missing resource now consistently return a
4xxerror (API errors) - API: ElevenLabs API keys and agent IDs are now checked with ElevenLabs when you save them. Invalid keys or inaccessible agents return
400right away instead of failing during a call (ElevenLabs Agent) - API: The error for deleting a resource that’s still in use now names the blocking agent next to its ID
- API: Starting a session or call on an avatar that is not available yet now returns
400right away instead of failing later - API: Outages of services the API depends on now return
503instead of500or403. Retry these with backoff (API Errors) - API: Sessions and conversations now start and end normally even when our payment provider is temporarily unavailable
Bug Fixes
- Fix avatar mouth moving slightly during silence
- Fix the avatar’s voice playing slightly ahead of its mouth movement at the start of some replies
- Fix agents with default Cartesia TTS re-cloning existing voices which needlessly increased conversation setup time
- Fix the conversation page crashing when you toggle a microphone or camera you don’t have
- Fix a custom avatar’s cloned voice occasionally being deleted after a conversation
- Fix occasional lip sync glitches when several sessions ran on the same server
- Fix avatar sessions occasionally dropping when another session’s room connection timed out
- Fix custom avatar creation failing for very short training videos (Custom avatars)
- Fix sessions with an ElevenLabs agent not ending when the ElevenLabs conversation drops
- Fix user messages occasionally being delivered twice in conversations
- Fix knowledge files sometimes not being indexed after upload (Knowledge Base)
- Fix new agents not appearing in the Studio agent list right after creation
- Fix Studio saving agents with avatars, knowledge files, or external APIs your organization can’t use
- Fix the upgrade plan button in Studio
- Fix links in avatar notification emails pointing to a missing page
- Fix sessions hanging instead of ending promptly when avatar rendering fails
- Fix agents with non-US Google voices (for example
en-IN-Neural2-B) speaking in a different voice instead of the configured one - Fix organizations with a past-due payment losing their plan’s limits before the subscription ends
- Fix plan changes occasionally not being applied, or being reverted, right after subscribing or upgrading
- Fix usage occasionally being counted against the wrong billing period length
- Fix large chunked uploads for avatars, knowledge files, and videos occasionally failing
Documentation
- Add ElevenLabs Agent integration guide
- Add External Meetings guide
- Add Text-to-Video documentation
- Add API Versioning and Deprecation pages
- Add a walkthrough video to the Speech-to-Video quickstart
- Add Studio avatar creation error messages to Troubleshooting
New Features
- Add self-service avatar creation in Studio (Custom Avatars)
- Add multi-factor authentication (MFA) for Studio accounts
- Add an invoices page in Studio settings with payment details and PDF downloads
- Add support for
cartesiaagent TTS configs - Add new region: asia-south1 (India)
Updates
- Avatars: Faster session startup thanks to avatar prefetching and parallel downloads
- Avatars: Reduced delay before the avatar starts speaking
- Agents: Changed default TTS to Cartesia Sonic 3.6 (from ElevenLabs Flash 2.5)
- Agents: Redesigned the conversation UI, including the pre-join and post-call screens
- Agents: Improved knowledge base handling to support larger files at lower latency
- Agents: Raised the knowledge file text limit from 100,000 to 250,000 characters (both in API and Studio)
- Studio: Redesigned the feedback flow to be less intrusive
- API: Renamed “Call” to “Conversation”:
/v1/callsis now/v1/conversations - API: Renamed
POST /v1/callstoPOST /v1/livekit-rooms(“Create LiveKit Room”) in a new “Agent Integration” group. You only need it when embedding a conversation via a LiveKit Client SDK - API: Old endpoints keep working but now return a
Deprecationheader with aLinkto their replacement
Bug Fixes
- Fix agents failing to start when a linked knowledge file was deleted
- Fix Google TTS for voices other than Chirp 3 HD
- Fix avatar audio occasionally stalling or skipping very short replies
- Fix avatar sessions not ending cleanly when the room closes before the agent joins
- Fix saving custom branding background URLs in Studio
- Fix incorrect avatar slot calculation; slots now correctly follow your plan (Concurrency & Quotas)
- Improve Studio sign-in error messages and fix a Google sign-in issue
Documentation
- Update Custom Avatars
- Add Knowledge Base guide
Breaking Changes
- Retired Genesis-1 video model: calls, sessions, and videos using a Genesis-1 avatar now return
410 Gone. All existing Genesis-1 avatars were already silently upgraded to Genesis-2.
Avatars
- Improve avatar lip sync by giving the model full audio history
- Improve session stability while avatar models are loading
Managed Agents
- Changed default OpenAI LLM to GPT 5.4 mini
- Declare AI interaction in the pre-join screen and in-call UI (following EU AI Act Art. 50.1)
- Fix room audio occasionally not being attached at the start of a call
Studio
- Added organizations with user management, role-based access control, and collaboration support
- Text-to-Video generation now includes visible watermarks (following EU AI Act Art. 50.4)
Documentation
- Redesign docs structure & navbar
- Redesign onboarding and “Get Started” section: Add Choose Path and quickstarts for Speech-to-Video and Managed Agents
- Update Roadmap
- Add Production Guides section with Concurrency & Quotas, GDPR & Privacy, EU AI Act, On-Prem Deployment, and Troubleshooting pages
- Document API error codes
- Add page context menu to copy a page or open it in an AI assistant
New Features
- Add avatar emotion control for more expressive, natural conversations
- Support agent creation with external / custom LLM config
- Support avatar creation via direct file upload
Studio
- Launch new revamped Studio experience with usage analytics, mobile support, and dark theme
Documentation
- Update API docs
New Features
- Add continuous vision agents for real-time perception
- Improve avatar idle animations
- Add message-logging control per agent
- Add billed-minutes usage API
Miscellaneous
- Add on-prem / enterprise deployment option
New Features
- Support custom avatar backgrounds
- Add custom user fields on calls in the developer API
Miscellaneous
- Improve API endpoint latency
New Features
- Add more speech-to-text (STT) providers for agents
- Add external / custom LLM support for agents
- Add Arabic (ar-SA) language support
- Add plan-based feature gating
Studio
- Add video generation support for Genesis-2 avatar videos
New Features
- Add Knowledge Base files API
- Add session status and start/end timestamps
Miscellaneous
- Serve downloads and previews via CDN
New Features
- Add session concurrency limits
- Add conversation transcripts via API
- Add ElevenLabs custom voice support
- Add Webhook events
Miscellaneous
- Add new region: europe-west1
New Features
- Add JavaScript support to LiveKit plugin
- Update API reference with Update Agent endpoint
- Update speech-to-video with Pipecat integration
Miscellaneous
- Embed demo as iframe in Getting Started
- Update stock avatars