ElevenAgents

Agent version metadata: Added an API endpoint to retrieve metadata for a specific agent version. The response includes the version ID, agent ID, branch ID, description, sequence number, commit timestamp, parents and access information.

Conversation filtering: Added a text_only filter to conversation list and message text search endpoints so developers can narrow results to text-only conversations.

Procedure loading: Added load_procedure as a system tool configuration with ProcedureAtVersion, and added skills to ProcedureCompilerMode. ProcedureSettings.compiler_mode now defaults to skills.

Agent configuration updates: Added gemini-3.1-flash-lite and qwen35-397b-a17b as LLM options, added genesys_bot_connector as a conversation initiation source, added auto_resolve_after_inactive_minutes to alerting settings, and added attributes_to_headers to SIP trunk configuration.

Custom LLM temperature: Agent prompt temperature can now be set to null to omit the temperature field from downstream LLM requests.

Workspaces

Auth connection updates: Added an API endpoint to update workspace auth connections. The endpoint supports updating basic auth, OAuth2 client credentials and OAuth2 JWT connections. OAuth2 client credentials now support custom_headers for token requests.

Auth connection types: Added bearer auth create request support and Slack bot auth response schemas.

Workspace resource types: Added convai_templates and transcription_tasks to workspace resource types.

Voice Isolator

Video processing status: Audio isolation history items now include the video_processing_failed boolean field.

SDK Releases

Python SDK

v2.48.0 - Regenerated the SDK for the May 18, 2026 API schema, including typed support for workspace auth connection updates, agent version metadata, conversation text_only filters, new ElevenAgents configuration fields, updated LLM options and voice isolator history updates. The release also excludes the speech_engine_custom.py example from Fern packaging.

JavaScript SDK

v2.48.0 - Added environment configuration to conversation initiation data and events, and regenerated the SDK for the May 18, 2026 API schema, including workspace auth connection updates, agent version metadata, conversation text_only filters, new ElevenAgents configuration fields, updated LLM options and voice isolator history updates.

Packages

@elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected] and @elevenlabs/[email protected] - Updated the client dependency. @elevenlabs/[email protected] pins livekit-client to 2.16.1 and forces the dual peer connection path to fix WebRTC connection failures with newer LiveKit join protocols.

@elevenlabs/[email protected] - Removed manually maintained types from @elevenlabs/types, including Role, Mode, Status, Callbacks, CALLBACK_KEYS, DisconnectionDetails, MessagePayload and AudioAlignmentEvent. Import those types from @elevenlabs/client instead. The types package now contains only generated code.

iOS SDK

v3.2.0 - Added support for text-only conversations over WebSockets and fixed message handling.

v3.1.5 - Improved ConnectionManager and DataChannelReceiver cleanup on deallocation, skipped audio hardware initialization in text-only mode, removed unused code and duplicate agent ID extraction, and bumped visionOS support.

API

Original source

ElevenAgents

SIP logs: SIP signaling logs are now available for SIP trunk calls in conversation history and phone number settings. Logs include the SIP call ID, phone numbers, addresses, transport, message direction, raw messages and errors to help debug call setup and routing issues.

Knowledge base document updates: New API to list indexed (RAG) chunks for a document. Documents can also be edited in place: rename the item and, for text-based documents, update the body without re-uploading the file. (File-based documents still use the existing replace-file flow where applicable.)

SMS conversation metadata: Added SMS support to conversation metadata with SMSConversationInfo, sms as an authorization method and twilio_sms as a conversation initiation source.

Agent configuration updates: Added background music configuration schemas, alerting monitor configuration, 2D layout fields, widget file upload configuration, and new LLM options gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-mini-2026-03-17 and gpt-5.4-nano-2026-03-17.

Workspaces

API request analytics: Added a workspace analytics endpoint for querying API requests with time range, filtering, search, sort and limit controls.

API key IP allowlisting: Added allowed_ips to service account API key create, edit and response schemas. Create requests accept an array of IP addresses or CIDR ranges, or null to allow all IPs. Edit requests also accept clear and no_update.

Workspace permissions: Added conversational_ai_read and voice_design to workspace group permissions.

Voice Design

Voice preview migration: Deprecated POST /v1/text-to-voice/create-previews. Use Design a voice to create previews, then use the returned generated_voice_id when creating a voice.

Voice Changer

WAV output formats: Added wav_8000, wav_16000, wav_22050, wav_24000, wav_32000, wav_44100 and wav_48000 to the output_format query parameter on speech-to-speech conversion.

SDK Releases

Python SDK

v2.47.0 - Regenerated the SDK for the May 12, 2026 API schema. The release adds typed support for RAG chunk listing, file document updates, workspace API request analytics, service account API key IP allowlisting, phone number agent configuration fields, webhook request headers, voice metadata moderation, new LLM options and updated speech-to-speech output formats.

JavaScript SDK

v2.47.0 - Regenerated the SDK for the May 12, 2026 API schema. The release adds typed support for RAG chunk listing, file document updates, workspace API request analytics, service account API key IP allowlisting, phone number agent configuration fields, webhook request headers, voice metadata moderation, new LLM options and updated speech-to-speech output formats.

Packages

@elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected] and @elevenlabs/[email protected] - Added full tool result payload support to onAgentToolResponse. The callback now receives agent_tool_response_full_payload events with the raw full_tool_result string, capped at 64 KB, when enabled in the agent configuration.

@elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected] and @elevenlabs/[email protected] - Added native mute and unmute support to Scribe realtime STT. RealtimeConnection now exposes mute(), unmute() and isMuted, and useScribe exposes matching state and callbacks. The releases also add onAgentResponseCorrection for agent response correction events.

@elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected] and @elevenlabs/[email protected] - Added optional contextId to sendContextualUpdate for deduplicating contextual updates and added llm to the typed agent prompt override for conversation sessions. The React package also allows useScribe microphone deviceId to use full ConstrainDOMString constraints.

@elevenlabs/[email protected] and @elevenlabs/[email protected] - Added file upload support to the embedded ElevenAgents widget.

@elevenlabs/[email protected] and @elevenlabs/[email protected] - Fixed text input submission for IME users by ignoring Enter keydowns while composition is active.

@elevenlabs/[email protected] and @elevenlabs/[email protected] - Updated the widget packages to @elevenlabs/[email protected], including full tool result payload support.

@elevenlabs/[email protected] and @elevenlabs/[email protected] - Fixed transcript ordering when user and agent messages share the same event_id, so voice and DTMF turns render the user transcript before the agent message.

@elevenlabs/[email protected] and @elevenlabs/[email protected] - Fixed voice widget transcripts so streamed agent response parts are ignored for voice sessions and late user transcripts are inserted before their matching agent response.

API

Original source

ElevenAgents

Conversation tags: Added first-class conversation tags for organizing and filtering conversations. You can now create, update, list and delete tags, assign tags to conversations, and filter conversation history by tag_ids.

Conversation list filters: Added exclude_statuses to list conversations, allowing clients to hide conversations with statuses such as initiated, in-progress, processing, done, or failed.

Model options: Added claude-opus-4-7, gpt-5.4, gpt-5.5, gpt-5.4-2026-03-05, gpt-5.5-2026-04-23, and qwen36-35b-a3b to the LLM enum.

Contextual update metadata: Added contextual_update_info to conversation transcript response items. The field references ContextualUpdateInfo, which includes context_id and is_superseded, so clients can identify contextual updates and whether they have been replaced by later context.

Batch calling filter: Added an optional agent_id query parameter to list workspace batch calling jobs, allowing clients to return jobs for a single agent.

Test invocation listing: Made the agent_id query parameter optional and nullable on list test invocations, allowing clients to list test invocations without filtering by one agent.

MCP response timeout: Increased the maximum response_timeout_secs value from 120 to 300 seconds on MCP server configuration and MCP tool configuration overrides.

Workspaces

Subscription overage: Added current_overage to SubscriptionResponseModel and ExtendedSubscriptionResponseModel.

Resource collections: Added resource_collection to the WorkspaceResourceType enum.

ElevenCreative Studio

Project image status: Added error and pending_task to ProjectImageResponseModel, and made signed image URLs nullable when an image is still processing or has failed.

Pending media fields: Removed legacy pending_block_ids and pending_external_audio_ids fields from project external audio and video response models.

SDK Releases

Python SDK

v2.46.0 - Added keyterms and no_verbatim support to the Scribe realtime API, refactored WebSocket URL construction, and regenerated the SDK for the May 7, 2026 API schema, including conversation tags, conversation list filters, batch calling agent_id filtering, optional test invocation agent_id, contextual update metadata, and MCP response timeout updates.

JavaScript SDK

v2.46.0 - Fern regeneration for the May 7, 2026 API schema, including conversation tags, conversation list filters, batch calling agent_id filtering, optional test invocation agent_id, contextual update metadata, and MCP response timeout updates.

Packages

@elevenlabs/[email protected] - Added keyterms (string[]) and noVerbatim (boolean) options to the Scribe realtime API. keyterms are sent as repeated WebSocket query parameters and can include up to 50 terms of up to 20 characters each. noVerbatim removes filler words, false starts, and disfluencies from transcripts. Also fixed a case where interruptions could cut off agent audio that arrived less than 2 seconds after the interruption.

@elevenlabs/[email protected] - Added shared types for Scribe realtime keyterms and noVerbatim.

@elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected], and @elevenlabs/[email protected] - Updated to the latest client package, including Scribe realtime options and the agent audio interruption fix.

@elevenlabs/[email protected] - Patch release that includes the Scribe realtime options, the agent audio interruption fix, and an automatic random user ID when none is set.

Android SDK

v0.9.0 - Added support for detecting user audio while the user is muted.

API

Original source

ElevenAgents

agent_response_complete WebSocket event: A new agent_response_complete client event fires exactly once when an agent's response is fully delivered -- after all LLM generation, tool call chains, and audio playback complete. The event lets clients reliably detect end of turn for batching messages, wake-word activation, or post-turn UI updates. The event is suppressed while async tool tasks are still running and is not emitted on agents configured with turn_timeout or soft_timeout. Added a new AGENT_RESPONSE_COMPLETE value to the ClientEvent enum and a corresponding AgentResponseCompleteClientEvent response model.

Pre-tool speech mode: Added a new pre_tool_speech field (PreToolSpeechMode enum, default auto) on MCP server configuration and per-tool overrides. Replaces the now-deprecated force_pre_tool_speech boolean. Values: auto decides whether the agent speaks before the tool call based on recent tool latency, force always asks the agent to speak, and off opts out entirely.

Tool dispatch respects execution mode: Workflow tool dispatch nodes now correctly respect each nested tool's execution_mode. If any tool in a dispatch node uses POST_TOOL_SPEECH, the workflow waits for the current turn's audio to be fully exposed before executing the dispatch.

MCP response timeout: Added a response_timeout_secs field (integer, default 30, min 5, max 120) on MCP server configuration and per-tool overrides, controlling the maximum time to wait for each MCP tool call to complete.

Agent trust context: Added a trust_context field on agent platform settings (AgentTrustContext enum: unknown | low | high, default unknown). low is for agents serving untrusted external participants, where outputs should be vetted and tool access scoped; high is for agents serving the owner where full tool access is appropriate; unknown covers agents created before this feature.

Source attribution toggle: Added a source_attribution boolean (default false) on conversation config. When enabled and knowledge base content is present, the LLM is instructed to report which sources it used.

Test sharing mode filter: The list tests endpoint now accepts a sharing_mode query parameter (all | shared_with_me, default all) to filter tests and folders to only those shared with the current user that they did not create.

Tool icons and execution metadata: Added icons (array of Icon) and execution (ToolExecution) fields to the Tool schema. Icon includes src, mimeType, and sizes for UI display, and ToolExecution.taskSupport (forbidden | optional | required) declares how the tool participates in tasks.

Procedure compiler default: The ProcedureCompilerMode default has changed from deterministic to append, and the llm value has been removed. Existing agents that explicitly set deterministic are unaffected.

RAG and knowledge base attribution metadata: Added used_static_kb_document_ids (array of strings) on ConversationHistoryTranscriptCommonModel and ConversationHistoryTranscriptResponseModel, and used_chunk_ids on RagRetrievalInfo, surfacing exactly which knowledge base documents and RAG chunks contributed to each transcript turn.

Dynamic variable refactor: dynamic_variables and dynamic_variable_placeholders now reference unified DynamicVariableValueType-Input / DynamicVariableValueType-Output schemas (replacing the prior DynamicVariableContainerValueType-* schemas), with broader support for nested and list values. DynamicVariableAssignment gained a preserve_native_type field to keep typed values rather than coercing to strings.

Audio Isolation

History endpoints: Two new endpoints expose your audio isolation history.

List audio isolation history (GET /v1/audio-isolation/history) returns a paginated list of all your audio isolation generations. Supports page_size, page, and search query parameters.

Delete audio isolation history item (DELETE /v1/audio-isolation/history/{history_item_id}) deletes a specific item and its associated media files.

Speech to Text

Realtime keyterms and verbatim mode: The Scribe realtime WebSocket now accepts a keyterms parameter (array of strings, max 50 entries of up to 20 characters each) to bias the model toward specific terms, and a no_verbatim boolean to remove filler words, false starts, and disfluencies from transcripts. Both parameters are echoed back in the session_started event. Available across the JavaScript SDK, Python SDK, and the @elevenlabs/client browser/Node client SDK.

SDK Releases

Python SDK

v2.45.0 - Fern regeneration for the April 27, 2026 API schema, including audio isolation history endpoints, tool executions, MCP pre_tool_speech and response_timeout_secs, and the agent_response_complete client event. Also includes a fix to the notebook install hint in the README.

v2.44.0 - Fern regeneration for the April 21, 2026 API schema, picking up the trust_context field, pre_tool_speech mode, asset transcription support, and MCP response timeout configuration.

JavaScript SDK

v2.45.0 - Added keyterms (string array, max 50 keyterms of up to 20 chars each) and noVerbatim (boolean) options on the Scribe realtime client. keyterms are sent as repeated query parameters on the WebSocket URL to bias transcription, and noVerbatim removes filler words and disfluencies. Both fields are also added to the typed Config interface for the session_started echo. Also includes the April 27, 2026 Fern regeneration (#376).

v2.44.0 - Fern regeneration for the April 21, 2026 API schema, picking up the trust_context field, pre_tool_speech mode, and MCP response timeout configuration.

Swift SDK

v3.1.4 - Fixed an issue where the local microphone kept publishing after an agent disconnected (silence timeout or end_call tool). onAgentDisconnected now disconnects the LiveKit room and tears down conversation state, with onDisconnect firing an .agent reason. ConnectionManaging.onAgentDisconnected is now async. Removed the unused ConversationConnectionManager wrapper.

Packages

@elevenlabs/[email protected] - Fixed the agent tool-call sound being interpreted as user speech and interrupting the agent during tool execution.

@elevenlabs/[email protected] - Added support for the upload conversation file endpoint, letting clients attach files to an in-progress conversation directly from the client SDK.

@elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to @elevenlabs/[email protected] / @elevenlabs/[email protected], picking up the tool-call sound fix and conversation file upload support.

@elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to the latest @elevenlabs/client for the same fixes.

@elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to @elevenlabs/[email protected] for the tool-call sound fix.

@elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to @elevenlabs/[email protected] for conversation file upload support.

API

Original source

Developer Resources

ElevenLabs Devs YouTube channel: A new developer-focused YouTube channel is now live, featuring tutorials, API walkthroughs, and technical deep dives to help you get the most out of the ElevenLabs platform.

ElevenAgents

Workflow starting node override: Conversations can now be started at a specific workflow node by passing starting_workflow_node_id in the conversation initiation client data. This is gated by an opt-in enable_starting_workflow_node_id_from_client field in the agent's platform override settings. When enabled, clients can specify which node to begin a conversation at, which is useful for agent memory and longer-term WebSocket use cases. If the specified node does not exist, the conversation falls back to the default entry point.

Post-call analysis LLM configuration: Agent platform settings now include an analysis_llm field (default: gemini-2.5-flash) that sets the default LLM for post-call analysis, including evaluation criteria and data collection. Individual data collection properties can override this default with a per-item llm field via the new AnalysisProperty schema, giving fine-grained control over which model processes each analysis item.

New LLM options: Added gpt-5.4 to the list of available LLM providers for agent configuration. Added xhigh to the LLMReasoningEffort enum for extended reasoning on supported models.

Null literal in workflow expressions: Workflow conditional expressions now support null comparisons via the new ASTNullNode type, enabling branches like "if variable is null" in workflow logic.

Get secret by ID: A new get secret endpoint (GET /v1/convai/secrets/{secret_id}) retrieves a single workspace secret by its ID. The list secrets endpoint also now accepts a search query parameter to filter secrets by name prefix.

Searchable phone numbers in batch calling: The batch calling form in the dashboard now includes a searchable phone number selector, making it easier to find the right outbound number when configuring batch calls.

Speech to Text

Speaker role detection: The convert speech to text endpoint now accepts a detect_speaker_roles parameter (boolean, default false). When enabled alongside diarize=true, speaker labels are returned as agent and customer instead of speaker_0, speaker_1. This cannot be used with use_multi_channel=true. Usage incurs an additional 10% surcharge on the base transcription cost.

Text to Dialogue

Input character limit guidance: The text to dialogue endpoints now document a recommended limit of 2,000 characters across all inputs[].text values per request. Longer requests may terminate early in streaming responses or return a validation error.

Workspaces

Usage analytics endpoint: A new usage by product over time endpoint (POST /v1/workspace/analytics/query/usage-by-product-over-time) returns credit usage broken down by product type over time. Results are returned in a tabular structure with configurable interval_seconds, group_by dimensions, and filters. The previous GET /v1/usage/character-stats endpoint has been deprecated in favor of this new endpoint.

SDK Releases

Packages

@elevenlabs/[email protected] - Fixed conversation startup readiness so onConnect runs after the session is marked connected and React has synchronized conversationRef. Also exposed and forwarded onConversationCreated for consumers that need the created Conversation instance before onConnect.

@elevenlabs/[email protected] - Fixed conversation startup readiness and exposed onConversationCreated callback, matching the @elevenlabs/[email protected] changes.

@elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to @elevenlabs/[email protected] with conversation startup fixes.

API

Original source

ElevenAgents

Conversation topic discovery: A new conversation topics endpoint (GET /v1/convai/agents/{agent_id}/topics) returns the latest topic discovery results for a given agent, surfacing recurring themes across conversations. The list conversations endpoint now accepts a topic_ids query parameter to filter conversations by discovered topic.

Knowledge base content search: A new search knowledge base endpoint (GET /v1/convai/knowledge-base/search) provides fuzzy text search over knowledge base document content. Results include highlighted snippets with SearchHighlightSegment objects and support cursor-based pagination with configurable page sizes (up to 100, default 30). Filter by document type with the optional types parameter.

Flexible branch merging: The merge branch endpoint no longer restricts merges to the main branch -- any branch can now be used as a merge target. A new optional force boolean parameter overrides timestamp-based conflict resolution when set to true. Branch metadata responses now include parent_branch_id and merged_from_version_id for better lineage tracking.

Secret dependency management: A new get secret dependencies endpoint (GET /v1/convai/secrets/{secret_id}/dependencies/{resource_type}) returns paginated dependency lookups filtered by resource type (tools, agents, or phone_numbers). The list secrets endpoint now accepts a dependency_limit query parameter to control how many dependencies are previewed per secret, with agents_has_more, tools_has_more, and phone_numbers_has_more flags indicating when additional dependents exist beyond the preview.

New LLM options: Added gemini-3.1-pro-preview, qwen35-35b-a3b, and qwen35-397b-a17b to the list of available LLM providers for agent configuration.

Voices

Non-community voice filter: The voice_type query parameter on the list voices endpoint now accepts non-community, which returns personal and workspace voices combined while excluding library copies. This provides a convenient filter for teams that want to see only their own voices without community-shared ones.

SDK Releases

Python SDK

v2.43.0 - Fern regeneration for the April 13, 2026 API schema, including knowledge base content search, conversation topics, flexible branch merging, and new LLM providers.

JavaScript SDK

v2.43.0 - Fern regeneration for the April 13, 2026 API schema, including knowledge base content search, conversation topics, flexible branch merging, and new LLM providers.

Swift SDK

v3.1.3 - Reduced background noise detection in software muted VAD, preventing false voice activity triggers when the microphone is muted.

Packages

@elevenlabs/[email protected] - Fixed getInputVolume() and getOutputVolume() returning 0 in React Native by adding native volume providers using LiveKit's RMS and multiband FFT processors. getInputByteFrequencyData() and getOutputByteFrequencyData() now return data focused on the human voice range (100-8000 Hz), which is more useful for voice visualization, and on web getInputVolume() and getOutputVolume() are also computed from this range.

@elevenlabs/[email protected] - Fixed startSession errors being swallowed instead of surfaced via onError in ConversationProvider. Previously, when Conversation.startSession() rejected (e.g., "agent not found"), the UI would get stuck in "connecting" with no error feedback.

@elevenlabs/[email protected] - Updated to @elevenlabs/[email protected] and @elevenlabs/[email protected] with native volume provider fixes.

@elevenlabs/[email protected] - Fixed transcript message ordering in voice mode where agent responses could appear before user messages.

API

Original source

ElevenAgents

• Scoped conversation analysis: Evaluation criteria and data collection items can now be scoped to conversation (full transcript) or agent (only the portion where a specific agent was active). Added scope field to PromptEvaluationCriteria, data_collection_scopes to agent platform settings, and a new scoped array of ScopedAnalysisResult on the conversation analysis response. This is particularly useful for multi-agent workflows where each agent should be evaluated independently.
• Agent test folder management: Tests can now be organized into folders. New endpoints for creating, retrieving, updating, and deleting test folders, plus bulk moving tests between folders.
• Tool and knowledge base overrides in workflows: Agent workflow node overrides now support tool_ids and knowledge_base fields in PromptAgentAPIModelOverrideConfig, allowing workflow nodes to control which tools and knowledge base documents each sub-agent can access.
• Response filter hide_all mode: Added hide_all option to ResponseFilterMode, forcing the system to hide all fields of a tool response from the agent.
• Visited agents in conversation history: The get conversation response now includes a visited_agents array of VisitedAgentRef objects (with agent_id and branch_id), tracking which agents participated in a multi-agent conversation.
• Multimodal message support in hooks: The useConversationControls hook now exposes sendMultimodalMessage, and the MultimodalMessageInput type is exported from @elevenlabs/client, making it easier to send images and other multimodal content during conversations.

Speech to Text

• Audio duration in responses: The convert speech to text response now includes an audio_duration_secs field, providing the total duration of the transcribed audio without requiring client-side calculation.

Voices

• Recording quality and review status: The get voice response now includes recording_quality (enum: studio, good, ok, poor, bad) and labelling_status (enum: in_review, review_complete) fields, providing visibility into voice quality assessment.

SDK Releases

JavaScript SDK

• v2.42.0 - Fern regeneration for the April 7, 2026 API schema, including scoped analysis, test folders, and DTMF input support.

Python SDK

• v2.42.0 - Fern regeneration for the April 7, 2026 API schema, including scoped analysis, test folders, and DTMF input support.

Swift SDK

• v3.1.2 - Fixed ObjC category dispatch for LiveKit delegate methods, resolving potential crashes in Swift-based voice agent integrations.

Packages

• @elevenlabs/[email protected] - Exposed sendMultimodalMessage in the client API and exported the MultimodalMessageInput type for sending images and other multimodal content during conversations.
• @elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected], @elevenlabs/[email protected] - Fixed Node.js ESM compatibility by adding explicit .js extensions to all relative imports and setting "type": "module" on @elevenlabs/types.
• @elevenlabs/[email protected] - Added automatic language selection from localStorage history and browser language preferences for the embeddable widget.

API

Original source

ElevenAgents

MCP tool scoping in agent workflows:

Agent workflow nodes can now restrict which MCP tools a sub-agent is permitted to call. When tool inheritance is disabled on a node, only the explicitly selected MCP tools are loaded for that sub-agent, giving teams precise control over tool access per workflow step.

Conversation file uploads:

The create agent and update agent endpoints now support a file_input field on ConversationConfig. When enabled, end users can attach images or PDFs in chat (requires an LLM with multimodal input support). Configurable with enabled (boolean) and max_files_per_conversation (integer).

Re-run conversation analysis:

New run conversation analysis endpoint (POST /v1/convai/conversations/{conversation_id}/analysis/run) re-evaluates a completed conversation using the agent's current evaluation criteria and data collection settings, without needing a new call.

Tool response mocking for tests:

Agent simulation tests and test suite invocations now support a tool_mock_config field to control how tool calls are handled during testing. Use MockingStrategy (all, selected, none) to choose which tools are mocked and MockNoMatchBehavior (call_real_tool, raise_error) to set the fallback when no mock matches.

Text search sort order:

The text search conversations endpoint now accepts a sort_by query parameter with values search_score (default) or created_at, allowing you to control whether results are ordered by relevance or recency.

mTLS auth connections:

Agent auth connections now support mutual TLS (mtls) as an auth_type, in addition to the existing options. New CreateMTLSAuthRequest and MTLSAuthResponse schemas are available for creating and retrieving mTLS-authenticated connections.

Max duration message:

Added max_conversation_duration_message field to agent configuration. When set to a non-empty string, the agent will send this message to the user when the maximum conversation duration is reached.

Branch and environment in conversation initiation:

Added branch_id and environment optional fields to the conversation initiation client data (ConversationInitiationClientDataRequest) and submit batch call request body, enabling routing to specific agent branches and environments.

Music

Video to music:

New POST /v1/music/video-to-music endpoint generates background music from one or more video files. Videos are combined in sequence. Accepts optional description (up to 1,000 characters) and tags (up to 10 style tags such as upbeat or cinematic) to influence the generated track.

Speech to Text

Transcribe from URL:

The convert speech to text endpoint now accepts a source_url parameter (string, optional) for transcribing audio or video from a hosted URL, including YouTube videos, TikTok videos, and other video hosting services. This can be used as an alternative to uploading a file directly.

Voices

Total count in shared voices list:

The list shared voices response now includes a total_count field, making it easier to implement pagination and display result counts.

SDK Releases

JavaScript SDK

• v2.41.0 - Added support for the multimodal_message WebSocket event type in ElevenAgents real-time conversations. Includes Fern regeneration for the latest API schema updates.
• v2.41.1 - Fern regeneration to match the April 1, 2026 API schema.

Python SDK

• v2.41.0 - Fixed audio_interface to be optional in text-only conversation mode, resolving a runtime error when starting a session without audio. Includes Fern regeneration for the latest API schema.

Packages

This release includes the new v1.0.0 of the client side Agent SDKs. It features major breaking changes to @elevenlabs/client, @elevenlabs/react, and @elevenlabs/react-native. Review the migration guidance below before upgrading.

To help upgrade, we released a Skill to help your agents upgrade for you. Install it with

npx skills add elevenlabs/packages

You can read more about the v1 release and its improvements on our developer blog.

@elevenlabs/[email protected] — Breaking changes:

Input and Output classes are no longer exported. Use the InputController and OutputController interfaces from @elevenlabs/client instead.

Conversation is no longer a class — it is now a namespace object and a type alias for TextConversation | VoiceConversation. Remove any instanceof Conversation checks and subclasses.

The default connectionType is now inferred from the conversation mode: voice conversations use "webrtc" by default, and text-only conversations use "websocket". To keep the previous behavior for voice, pass connectionType: "websocket" explicitly.

VoiceConversation.wakeLock is now private. Pass useWakeLock: false in session options to suppress wake lock management.

changeInputDevice() and changeOutputDevice() now return Promise<void> instead of Promise<Input> or Promise<Output>.

Replace conversation.input.analyser.getByteFrequencyData(data) with conversation.getInputByteFrequencyData().

Replace conversation.input.setMuted(v) with conversation.setMicMuted(v).

Replace conversation.output.gain.gain.value = v with conversation.setVolume({ volume: v }).

getInputVolume(), getOutputVolume(), getInputByteFrequencyData(), and getOutputByteFrequencyData() now return 0 or an empty Uint8Array instead of throwing when no conversation is active.

@elevenlabs/[email protected] — Breaking changes:

useConversation now requires a ConversationProvider ancestor. Wrap your component tree in <ConversationProvider> and move options to the provider or to the hook.

DeviceFormatConfig and DeviceInputConfig exports are removed. Use FormatConfig and InputDeviceConfig from @elevenlabs/client instead.

New granular hooks replace the monolithic useConversation for better render performance: useConversationControls(), useConversationStatus(), useConversationInput(), useConversationMode(), useConversationFeedback(), and useRawConversation(). Each hook subscribes only to the state it needs, preventing unnecessary re-renders.

New useConversationClientTool(name, handler) hook for registering client tools that agents can invoke, with automatic cleanup on unmount.

Added controlled mute support via isMuted and onMutedChange props on ConversationProvider.

@elevenlabs/[email protected] — Breaking changes:

The previous ElevenLabsProvider and useConversation API have been removed and replaced with re-exports from @elevenlabs/react. Replace ElevenLabsProvider with ConversationProvider and useConversation with the granular hooks (useConversationControls, useConversationStatus, etc.).

On React Native, the package now polyfills WebRTC globals, configures the native AudioSession, and registers a platform-specific voice session strategy on import.

@elevenlabs/[email protected]

• Exports the CALLBACK_KEYS runtime array containing all keys from the Callbacks interface, used internally by the React SDK for callback composition.

@elevenlabs/[email protected]

• Added the guardrail_triggered server-to-client WebSocket event and the onGuardrailTriggered callback, which fires when the server detects a guardrail violation during a conversation. Also added type discriminants to TextConversation and VoiceConversation to enable discriminated union narrowing, and added startSession overloads that narrow the return type based on the textOnly option.

@elevenlabs/[email protected]

• Added the guardrail_triggered WebSocket event and onGuardrailTriggered callback, consistent with @elevenlabs/[email protected].

@elevenlabs/[email protected]

• Added client-side support for mocking tool responses in agent conversations, enabling test scenarios that simulate tool call outcomes without invoking real tools.

@elevenlabs/[email protected]

• Added type definitions for tool response mocking in agent conversations.

@elevenlabs/[email protected]

• Updated to depend on @elevenlabs/[email protected].

@elevenlabs/[email protected]

• Updated to depend on @elevenlabs/[email protected] and @elevenlabs/[email protected].

API

Original source

ElevenAgents

Environment Variables API: New environment variables endpoints for managing workspace-level configuration that agents can reference at runtime via {{system_env__<label>}} templating. Supports string, secret, and auth-connection variable types with per-environment value overrides. Endpoints include create, list, get, and update.

Auth Connections management: New workspace-level auth connections API for managing authentication credentials used by agent tools and integrations. Supports multiple auth methods including OAuth2 client credentials, JWT, basic auth, bearer tokens, custom headers, and integration-managed OAuth2 authorization code flows. Endpoints include create, list, and delete operations.

Knowledge Base URL refresh: New refresh endpoint (POST /v1/convai/knowledge-base/{documentation_id}/refresh) to re-fetch and update content for URL-sourced knowledge base documents. Knowledge base documents also now support auto-sync configuration with enable_auto_sync, auto_remove, and auto_sync_info fields.

Guardrail retry with feedback: Custom and content guardrails now support configurable trigger actions with EndCallTriggerAction and RetryTriggerAction options. When set to retry, the agent re-generates its response with injected system feedback up to 3 attempts. Available placeholders include {{trigger_reason}} and {{agent_message}} for contextual retry guidance.

Conversation History system variable: Added a new system__conversation_history dynamic variable that provides a lazily-evaluated, JSON-serialized conversation history at runtime. This is useful for passing full conversation context to tools, webhooks, or sub-agent handoffs.

Webhook tool content type: Server tools now support a configurable content type for webhook body parameters, allowing you to choose between application/json and application/x-www-form-urlencoded formats. The URL-encoded format is useful for integrating with legacy systems, OAuth token endpoints, and payment processors.

WhatsApp outbound messages: Updated the WhatsApp integration with a new outbound message dialog in the dashboard, enabling agents to send outbound messages in addition to calls through the WhatsApp channel.

Agent and resource listing filters: The list agents, list knowledge base documents, and list tools endpoints now support a created_by_user_id query parameter (use @me for the current user). The previous show_only_owned_agents and show_only_owned_documents parameters are deprecated.

Workflow conditional expressions: Added a new conditional_operator AST node to the workflow expression schema, enabling branching logic within agent workflow definitions.

Music

Music Marketplace: New Music Marketplace documentation covering the marketplace for licensing AI-generated music, including usage types, creator payouts, and licensing details.

SDK Releases

Python SDK

v2.40.0 - Added environment parameter support for ElevenAgents conversations, enabling environment-specific agent connections. Fern regeneration to match the latest API schema including environment variables, auth connections, knowledge base refresh, and guardrail trigger actions.

JavaScript SDK

v2.40.0 - Added multimodal_message WebSocket event type for ElevenAgents real-time conversations. Fern regeneration to match the latest API schema including environment variables, auth connections, knowledge base refresh, and guardrail trigger actions.

iOS SDK

v3.1.1 - Added environment parameter support for environment-specific agent connections. Fixed a visionOS build error by bumping the platform requirement to v2.

Packages

@elevenlabs/[email protected] - Added multimodal_message WebSocket event type for ElevenAgents real-time conversations.

@elevenlabs/[email protected] - Added multimodal_message WebSocket event support.

@elevenlabs/[email protected] - Added multimodal_message WebSocket event support.

@elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to @elevenlabs/[email protected].

@elevenlabs/[email protected], @elevenlabs/[email protected] - Updated to @elevenlabs/[email protected].

Packages v1.0.0 Release Candidate

The first release candidate for v1.0.0 of the ElevenAgents client SDKs is now available. This is a major release with breaking changes that improve the API surface, add granular React hooks for better render performance, and unify the React Native SDK with the React SDK.

To try out the release candidate:

npm install @elevenlabs/client@next @elevenlabs/react@next @elevenlabs/react-native@next

An elevenlabs:sdk-migration skill is available to help AI coding assistants automatically migrate your codebase to the new APIs. Add it to your Claude Code, Cursor, or Windsurf project to get guided migration support.

Key changes in v1.0.0:

@elevenlabs/client: Conversation is now a namespace object and type alias for TextConversation | VoiceConversation instead of a class. The Input and Output classes are replaced by InputController and OutputController interfaces, with new convenience methods on the conversation instance (setMicMuted(), setVolume(), getInputByteFrequencyData(), getOutputByteFrequencyData()). changeInputDevice() and changeOutputDevice() now return void.

@elevenlabs/react: useConversation now requires a ConversationProvider ancestor. New granular hooks for fine-grained re-rendering: useConversationControls(), useConversationStatus(), useConversationInput(), useConversationMode(), useConversationFeedback(). New useConversationClientTool() hook for dynamically registering client tools from React components with full type safety.

@elevenlabs/react-native: Complete API rewrite replacing ElevenLabsProvider and useConversation with ConversationProvider and the same granular hooks from @elevenlabs/react. WebRTC polyfills and native AudioSession configuration are applied automatically on import.

See the full release notes and migration guides:

@elevenlabs/[email protected]

@elevenlabs/[email protected]

@elevenlabs/[email protected]

@elevenlabs/[email protected]

API

Original source

ElevenAgents

Users page is now generally available: The list users page, which groups conversations by a user identifier, is now available to all workspaces. This allows you to view all users who have interacted with your agents, their conversation history, and contact details in a unified view.

SIP inbound headers as dynamic variables: Custom SIP X- headers from inbound SIP trunking calls are now automatically exposed as dynamic variables in ElevenAgents conversations. Any custom SIP header (e.g., X-Contact-ID, X-Campaign-ID) passed by the caller is available in the agent prompt using {{sip_contact_id}}, {{sip_campaign_id}}, etc. These variables are also visible in the conversation history under the Phone Call tab. Reserved headers such as X-Call-ID and X-Caller-ID continue to map to system__call_sid and system__caller_id and are not overridden.

Conversation filtering by tool outcome: The list conversations and text search conversations endpoints now accept tool_names_successful and tool_names_errored query parameters (array of strings) to filter conversations by which tools succeeded or returned errors during the call.

User listing improvements: The list users endpoint now supports a sort_by query parameter accepting last_contact_unix_secs (default) or conversation_count, and a branch_id query parameter to filter users by agent branch.

Force delete tools: The delete tool endpoint now accepts a force query parameter (boolean, default false). When set to true, the tool is deleted even if it is used by agents, and it is automatically removed from all dependent agents and branches.

Conversation embedding retention: The ElevenAgents settings response now includes conversation_embedding_retention_days, which controls how long conversation embeddings are retained (maximum 365 days). A null value uses the system default of 30 days.

Content threshold guardrail: Added the ContentThresholdGuardrail schema, which provides a configurable threshold-based guardrail with an is_enabled flag and a threshold value for content moderation.

Workspaces

Get all workspace groups: New GET /v1/workspace/groups endpoint returns all groups in the workspace, including each group's name, ID, members, permissions, usage limit, and character count.

Seat type in bulk workspace invites: The invite multiple users endpoint now accepts an optional seat_type field to specify the seat type (e.g., workspace_member, workspace_admin) for all invited users in a bulk operation.

Mobile SSO reliability improvements: Fixed a regression where SSO login on mobile would get stuck on "Authenticating..." after a user logged out and attempted to sign in again. Also fixed workspace switching for mobile SSO users who sign in via SAML or OIDC providers.

Music

Section duration control: The generate music detailed endpoint now supports respect_sections_durations (boolean), which controls whether the model adheres to the duration specified for each section in the composition plan.

Studio

Chapter visual content indicator: Chapter response models now include a has_visual_content boolean field indicating whether the chapter contains visual content.

Text shadow and outline styles: Studio text styling now supports text_shadow and text_outline options via the new StudioTextStyleShadowModel and StudioTextStyleOutlineModel schemas, enabling richer visual text customization within Studio projects.

Media generation clip task type: The PendingClipTask.type enum now includes the media_generation value, expanding clip task categorization to cover media generation workflows.

SDK Releases

Python SDK

• v2.39.0 - Added support for the multimodal_message WebSocket event type in ElevenAgents real-time conversations.
• v2.39.1 - Fern regeneration to match the latest API schema.

JavaScript SDK

• v2.39.0 - Fern regeneration to match the latest API schema.

iOS SDK

• v3.1.0 - Several improvements to the ElevenAgents Swift SDK:Server errors are now surfaced through the onError callback, making error handling more consistent.
• Added event-based agent state management for cleaner state observation.
• Software muting is now supported alongside speech detection, enabling more granular audio control.
• Improved codebase to use Swift Concurrency throughout.
• Fixed a crash that occurred during startup on iOS release builds.
• Updated branding to reflect the ElevenAgents rebrand.

Packages

• @elevenlabs/[email protected] - Added audio resampling for output audio when the required sample rate cannot be acquired from the system, improving compatibility across devices.
• @elevenlabs/[email protected] - Updated to depend on @elevenlabs/[email protected].
• @elevenlabs/[email protected] - Updated to depend on @elevenlabs/[email protected].

API

Original source

ElevenAgents

New LLM options: Added claude-sonnet-4-6 and gemini-3.1-flash-lite-preview as supported LLM providers for agent conversation configuration.

Guardrail execution mode: Custom guardrails now support a configurable execution_mode field with the GuardrailExecutionMode enum, and have separate input and output schema definitions. This allows more precise control over when guardrail logic runs and which message direction it applies to.

Guardrail triggered client event: Added guardrail_triggered to the ClientEvent enum, enabling agent workflows and client-side handlers to respond when a guardrail fires during a conversation.

Batch calling concurrency: Added target_concurrency_limit field to the submit batch call request body, allowing you to control the target number of concurrent calls for a given batch.

Agent testing type filter: The list agent tests endpoint now accepts a types query parameter to filter results by test type. The include_folders query parameter is deprecated in favor of the new types filter.

Telephony call configuration: Added telephony_call_config to outbound call request schemas for Twilio and SIP trunk calls, with a default ringing_timeout_secs of 60 seconds.

WhatsApp audio message responses: Added enable_audio_message_response to WhatsApp account configuration, enabling agents to send audio messages in addition to text in WhatsApp conversations.

Studio

Studio agent settings: Studio project models now expose an agent_settings field using StudioAgentSettingsModel, giving agent-connected studio projects a consistent settings surface alongside the new StudioAgentToolSettingsModel.

Mobile

SAML SSO for iOS and Android: Enterprise customers using SAML-based identity providers can now sign in from the ElevenLabs iOS and Android apps.

Speech to Text

No verbatim transcription mode: Added no_verbatim as a transcription option on the convert speech to text endpoint, allowing the model to produce cleaned-up output rather than a verbatim transcript.

Music

SDK CRLF fix: Resolved an issue in the Python and JavaScript SDKs where the generate music detailed endpoint response was not correctly parsed when the server used \r\n\r\n line endings. The parsers now handle both \r\n and \n line endings.

Voices

Voice captcha font support for non-Latin scripts: Fixed an issue where the instant voice cloning captcha displayed unreadable characters for some non-Latin scripts. Added font support and correct language selection for Hebrew, Thai, Armenian, Georgian, Malayalam, Telugu, and Gurmukhi. The captcha now uses the actual language of the cloned voice rather than defaulting to English.

SDK Releases

JavaScript SDK

• v2.38.0 - Fern regeneration to match the latest API schema, including new ElevenAgents fields, updated SMB tool types, and studio agent settings.
• v2.38.1 - Fixed CRLF parsing for the music detailed endpoint response stream.

Python SDK

• v2.38.0 - Fern regeneration to match the latest API schema, including new ElevenAgents fields, updated SMB tool types, and studio agent settings.
• v2.38.1 - Fixed CRLF parsing for the music detailed endpoint response stream.

Packages

@elevenlabs/[email protected] - Fixed a widget crash on Wix sites where Wix security hardening makes addEventListener non-writable.
@elevenlabs/[email protected] - Registered missing livekit-client pnpm patch in patchedDependencies.
@elevenlabs/[email protected] - Fixed a widget crash on Wix sites where addEventListener is non-writable.
@elevenlabs/[email protected] - Registered missing livekit-client pnpm patch in patchedDependencies.

API

Original source

ElevenAgents

New widget configuration options: Several new settings are now available for the ElevenAgents conversation widget in the agent dashboard:
Collapsible widget (widget.dismissible): Allow users to minimize or dismiss the widget during a conversation.
Action indicator (widget.show_agent_status): Display a visual indicator when the agent is actively using tools, with distinct states for working, done, and error. Enabling this setting automatically adds agent_tool_request and agent_tool_response to the agent's client events.
Conversation ID display (widget.show_conversation_id): Show the conversation ID to users after a call ends. Defaults to true for both new and existing agents.
Audio tag visibility (widget.strip_audio_tags): Hide audio tags from conversation transcripts. Defaults to true for both new and existing agents.
Syntax highlighting theme (widget.syntax_highlight_theme): Configure code block highlighting in transcripts. Accepts null (Auto), light, or dark.
Folder-aware agent testing: The List agent tests endpoint now supports organizing tests into folders. Added parent_folder_id, include_folders, and sort_mode query parameters for filtered listing. Test creation and update endpoints accept parent_folder_id to assign tests to folders, and test summaries now include entity type, parent folder path, and children count.
Workflow say node: Added a new say node type to agent workflows (WorkflowSayNodeModel). This node supports conversation_config, additional_prompt, and tool and knowledge base overrides. The node's message payload uses a discriminated union with two variants: literal (static text via text) and prompt (LLM-generated via prompt).
SMB tool configuration: Formalized the SMBToolConfig schema with a required, discriminated params object. Supported operation types are create, list, search, update, and delete, each applicable to SMB entity categories including clients, staff, services, products, and assets.
Summary language for agents: Added summary_language field to agent configuration to specify the language for post-conversation summaries.
Twilio call recording: Added optional call_recording_enabled field to the Twilio outbound call request body to enable recording of outbound calls.
WhatsApp messaging flag: Added enable_messaging flag to WhatsApp account models and update requests to control whether messaging is enabled for a given phone number.

Pronunciation Dictionaries

Set all rules at once: New Set rules endpoint (POST /v1/pronunciation-dictionaries/{pronunciation_dictionary_id}/set-rules) replaces all existing rules in a pronunciation dictionary in a single call, as an alternative to incrementally adding or removing individual rules.
Rule matching options: Added case_sensitive (boolean) and word_boundaries (boolean) fields to pronunciation dictionary rule definitions for more precise control over how rules match text.

Music

Upload audio: New Upload music endpoint (POST /v1/music/upload) accepts a multipart/form-data request with an audio file. Optionally extracts a composition plan from the uploaded audio and returns a MusicUploadResponse with a song_id.
Phonetic name support: Added use_phonetic_names parameter to Generate music, Generate music detailed, and Stream music endpoints.
Song ID in responses: Music generation responses now surface the song_id field, making it easier to reference generated songs in subsequent API calls.

Audio Native

Update project content from URL: New Update content endpoint (POST /v1/audio-native/content) allows updating an AudioNative project by providing a URL. The endpoint extracts the content from the URL and queues it for conversion and auto-publishing.

Voices

Voice bookmarking: Added bookmarked field to voice update requests and is_bookmarked to voice response objects, enabling users to bookmark voices for easy access.

SDK Releases

JavaScript SDK

v2.37.0 - Added support for surfacing song_id in music generation responses. Updated to include latest API schema changes from Fern regeneration.

Python SDK

v2.37.0 - Added support for new music generation parameters and surfacing song_id in music generation responses. Updated to include latest API schema changes from Fern regeneration.

Packages

@elevenlabs/[email protected] - Propagated event_id through transcript and streaming callbacks. Refactored tool status tracking from Map-based to inline transcript entries with a display-transcript utility. Added show-conversation-id config option (boolean, defaults to true) to control visibility of conversation ID in disconnection messages.
@elevenlabs/[email protected] - Propagated event_id through transcript and streaming callbacks. Refactored tool status tracking to inline transcript entries.
@elevenlabs/[email protected] - Propagated event_id through transcript and streaming callbacks.

API

Original source

ElevenAgents

OAuth and advanced authentication for MCP servers: MCP servers now support workspace auth connections, enabling OAuth2 Client Credentials, Basic Auth, Bearer Auth, JWT, and custom header authentication. Select an auth connection when creating or editing an MCP server to automatically handle token refresh and authentication headers. See the MCP server documentation for more details.

LLM information endpoint: Added a new endpoint to list available LLMs with deprecation status, capabilities, and context limits. This enables clients to programmatically discover which models are available and which are being deprecated. See List Available LLMs for details.

Conversation history redaction: Added support for redacting sensitive information from conversation transcripts, audio, and analysis before being stored. Configure specific entity types to redact (such as names, email addresses, etc.) using the new conversation_history_redaction setting in agent privacy configuration. See the docs page for more details.

Enhanced guardrails: Introduced two new guardrail types for agents:

• focus: Helps keep conversations on-topic
• prompt_injection: Detects and prevents prompt injection attempts
  Note: The alignment guardrail has been removed

Conversation search endpoints: Added two new endpoints for searching conversation messages:

• Text search - Full-text and fuzzy search over transcript messages
• Smart search - Semantic search using embeddings

File uploads in conversations: Added endpoints to upload and manage files within conversations:

• Upload file - Upload files to a conversation
• Delete file - Remove uploaded files

New embedding model: Added support for qwen3_embedding_4b embedding model in knowledge base RAG indexing.

Workspaces

Image and video generation permission: Workspace administrators can now control access to image and video generation features. This new permission allows you to restrict which workspace members can use these capabilities. Configure this in your workspace settings under member permissions.

Music

Workspace sharing: Songs can now be shared within workspaces. The songs resource type has been added to workspace resource sharing endpoints.

SDK Releases

JavaScript SDK

v2.36.0 - Added overloaded convert method signatures to Speech to Text wrapper for improved type safety and ergonomics. Updated SDK to include latest API schema changes including LLM list endpoint, conversation search, and MCP auth connection support.

Python SDK

v2.36.1 - Added missing music generation parameters including seed, loudness, quality, and guidance_scale to ensure full feature parity with the API.
v2.36.0 - Renamed package references from "Conversational AI" to "ElevenAgents" to reflect the product rebrand. Updated SDK to include latest API schema changes including LLM list endpoint, conversation search, and MCP auth connection support.

Packages

• @elevenlabs/[email protected] - Updated widget core with latest features and improvements
• @elevenlabs/[email protected] - Updated embeddable widget with latest features and improvements
• @elevenlabs/[email protected] - React Native SDK updates
• @elevenlabs/[email protected] - Updated TypeScript types package

API

Original source

ElevenAgents

• Conversation users endpoint: Added Get conversation users endpoint (GET /v1/convai/users) to list users who have had conversations with your agents. Supports pagination and filtering by agent, time range, and other criteria.

• Agent versioning: Fetch specific agent configurations by version using the new version_id and branch_id query parameters on the Get agent endpoint. The Update agent endpoint also now accepts a branch_id parameter for branch-specific updates.

• Search documentation tool: Added search_documentation as a new built-in system tool for RAG (retrieval-augmented generation). Configure multi-source retrieval with MultiSourceConfigJson, SourceConfigJson, and SourceRetrievalConfig schemas. Supports configurable merging strategies via the MergingStrategy enum.

• MCP tool support: Added mcp as a new tool type alongside webhook, client, and system tools. Configure MCP tools using the new MCPToolConfig schema in your agent tool definitions.

• Content guardrails: Added content moderation guardrails to GuardrailsV1 with configurable thresholds for different content categories: sexual, violence, harassment, self-harm, profanity, religion/politics, and medical/legal content. Use the new ContentGuardrail and ContentConfig schemas.

• Expressive mode: Added expressive_mode field (boolean) to agent configuration schemas to automatically prompt your agent to make the most of the new v3 conversational model.

• Post-dial digits: Phone number transfers now support post_dial_digits for sending DTMF tones after connection. Configurable as static values or dynamic variables.

• Agent testing types: Agent testing now supports three distinct test types with dedicated schemas: llm (response evaluation), tool (tool-call verification), and simulation (full conversation simulation). Simulation tests include simulation_scenario and simulation_max_turns configuration fields.

• Pronunciation Dictionaries

• Rules in API response: The Get pronunciation dictionary endpoint now returns a rules array containing full rule details. Response uses the new GetPronunciationDictionaryWithRulesResponseModel schema with PronunciationDictionaryAliasRuleResponseModel and PronunciationDictionaryPhonemeRuleResponseModel for rule types.

• Knowledge Base

• Folder deletion: The Delete knowledge base document endpoint now supports deleting folders in addition to documents. When deleting folders, set force=true to enable recursive deletion of all contained documents and subfolders.

• SDK Releases

• Python SDK

• v2.36.0 - Added conversation users endpoint, agent versioning, search documentation tool, MCP tool support, content guardrails, and agent testing types

• JavaScript SDK

• v2.36.0 - Added overloaded convert signatures to speechToText.convert() for improved type inference based on request parameters (access .text, .transcripts, or webhook fields without manual type narrowing), added conversation users endpoint, agent versioning, search documentation tool, MCP tool support, content guardrails, and agent testing types

• Widget Packages

• @elevenlabs/[email protected] - Added agent tool usage status display and new status badge for long-running tool calls, fixed emotion tag stripping, fixed rating and feedback submission for signed-url widget embedding

• @elevenlabs/[email protected] - Added agent tool usage status display and new status badge for long-running tool calls, fixed emotion tag stripping

• API

Original source