> ## Documentation Index > Fetch the complete documentation index at: https://docs.interactly.ai/llms.txt > Use this file to discover all available pages before exploring further. # Product updates > New releases and improvements ## What's new in v1.17.7 πŸš€ Updates & Improvements ### πŸ€– AI Models & Voice * **Broader Gemini support** β€” Added Gemini 3.5 Flash and additional real-time/Audio model variants, improved model selection logic, thought-signature tracking, and assistant thinking-budget configuration (including a fix for thinking-budget not saving on new assistants). * **Native streaming improvements** β€” Rolled out native response streaming for more model providers with a proactive fallback path, plus sequential LLM fallback groups for more resilient model selection. * **Named LLM configurations** β€” Teams can create and manage reusable, named LLM configs with team-level defaults and per-node overrides. *** ### πŸŽ™οΈ Voice, TTS & Telephony * **Telnyx warm transfer & call forwarding** β€” Telnyx calls now support warm transfers and call forwarding, alongside workflow-level call-forwarding detection. * **New Deepgram voices** β€” Added TTS voices for more languages, a specialized medical STT model for self-hosted transcription, and adjustable Aura 2 speech speed (0.7x–1.5x) for English and Spanish. * **Streaming reliability fixes** β€” Smoother call termination and barge-in handling mid-stream, fixed background sound playback during streamed calls, and resolved Vonage audio streaming issues. * **Updated ElevenLabs & Cartesia** β€” Improved ElevenLabs streaming reliability with concurrency tracking, and an updated default Cartesia voice model for better quality. *** ### πŸ”„ Workflows * **Favorites & filters** β€” Mark workflows as favorites and filter/search by creator, favorites, or workflow ID. * **More reliable conditions** β€” Fixed condition expressions with quotes evaluating incorrectly. * **Runtime language variable** β€” Workflow nodes can reference the conversation's language at runtime. * **Better diagnostics** β€” Improved tracking for calls that end or error out mid-workflow. *** ### πŸ₯ Integrations (EHR / Scheduling) * **Scheduling accuracy fixes** β€” Resolved several appointment scheduling/rescheduling issues (date/time handling, patient gender field, coverage eligibility, slot references). * **Patient appointments support** β€” Added EHR gateway support for retrieving patient appointments directly. * **Richer coverage details** β€” Coverage lookups now include payor name and additional plan details. * **Faster imports** β€” Bulk integrations imports now run in the background to prevent timeouts on large imports. * **Scheduling page** β€” The web scheduling flow now displays privacy policy and terms of service links. *** ### πŸ’¬ SMS * **New SMS sending API** β€” Added a public API for sending SMS messages, with rate limiting to prevent abuse. * **Log lookup & tracking** β€” Retrieve SMS logs by ID, with improved tracking for inbound and outbound messages, and SMS logs now configurable as a user role feature. *** ### πŸ“£ Campaigns * **Soft delete** β€” Deleting a campaign or call now soft-deletes it; hard deletion is restricted to admins. * **Pause policy groundwork** β€” Added evaluation logic for automatically pausing campaigns based on policy rules. * **Removed auto-timeout** β€” Campaigns no longer time out automatically after a few hours. * **Enhanced webhooks** β€” Improved webhook support for campaign events. * **Cloning fix** β€” Cloning a campaign no longer fails when a source call has a missing or empty outcome. *** ### πŸ“Š Analytics & Usage * **Organization usage stats** β€” Added monthly per-organization usage tracking (calls and minutes), with a read API and scheduled reporting. * **Outcome-based filtering** β€” Conversations can now be filtered by empty/null/missing outcome, outcome stats support more filter options, and a new endpoint returns conversation stats grouped by outcome. * **Audit log statistics** β€” Added new statistics endpoints for audit logs. * **More accurate call data** β€” Improved end-of-call metadata capture for accurate outcomes, fixed incorrect call duration in stats, and added call duration/cost stats at the assistant level. * **Webhook visibility** β€” Added APIs to view webhook delivery stats and latency. * **Call end attribution** β€” Call end events now record who or what ended the call. *** ### πŸ› Bug Fixes & Platform * Fixed a call-pricing issue where missing data could cause errors in cost calculations. * Reduced server startup CPU load and optimized several backend queries (tag-based stats, workflow run cleanup). * Fixed a communication preferences issue, an internal ID-handling bug in the admin service, and a caching serialization issue. * Reduced unnecessary text-to-speech calls for silent responses, improving latency. * Added TTS streaming performance metrics. ## What's new in v1.16.6 πŸš€ Updates & Improvements ### πŸ”„ Workflows β€” Now Live for Assistant Calls Workflows are now generally available for assistant-driven voice calls. You can build multi-step, branching conversation flows that execute deterministically alongside live calls β€” replacing single-prompt assistants with structured, auditable pipelines. #### What's new in the workflow engine * **Conditional Exit Edges & Global Routing** β€” Super nodes now support conditional exit edges, reverse edges, global routing, and guardrail propagation for fine-grained control over conversation branching and escalation paths. * **Guardrail Strike System** β€” Automatic guardrail-strike counting with configurable escalation thresholds and auto-finish behavior prevents runaway conversations. * **Resume from Any Node** β€” Workflows can be resumed from any node with an initial state context, enabling retry and recovery flows without a full restart. * **Workflow Lock** β€” Lock a workflow to prevent concurrent modification while it is in active use. * **Field Extractor Node** β€” Extract structured fields from unstructured data using the new Field Extractor node (Data Transformation category). * **Deduplicate Node** β€” Remove duplicate records in data pipelines using the new Deduplicate node (Data Extraction category). * **Dynamic Messages on Conditional Edges** β€” Conditional edge transitions can trigger dynamic narration messages for smoother conversational transitions. * **Warm-Transfer Context** β€” Transferred calls now receive full conversation history and runtime tool hydration. * **Stateful REST Execution** β€” A stateful turn-by-turn REST endpoint supports non-voice (REST-driven) workflow execution. * **Multi-Version Export/Import** β€” Workflows can be exported and imported as versioned bundles for backup, migration, and deployment. * **Workflow Run Metrics** β€” LLM latency, token usage, and operational metrics are captured per workflow iteration and per parallel condition evaluator. *** ## Bandwidth Deprecation Notice **Deprecation Notice**: Bandwidth telephony support will be deprecated in a future release. Please migrate your numbers to Twilio, Vonage, or Telnyx. *** ### πŸ“ž Telnyx Telephony Support Telnyx is now a fully supported telephony provider for inbound and outbound voice calls. * **Inbound & Outbound Calls** β€” Full call handling via Telnyx TeXML applications. * **Call Recording** β€” Recordings are stored automatically for Telnyx calls. * **Number Management** β€” Import Telnyx phone numbers and manage call hooks directly from the dashboard. *** ### πŸ” Twilio Warm Transfer (Human Agent Handoff) Outbound calls forwarded via Twilio now support **warm transfer** β€” a three-way handoff that lets an agent speak with the human recipient before the caller is connected. * Bi-directional context passing so the receiving agent has full conversation history. * Configurable timeouts and caller ID fixes. * Welcome messages are suppressed during warm transfer to avoid confusing the receiving party. * Calls can be recorded through the entire warm-transfer flow. *** ### πŸ“Š Assistant Analysis β€” DND & Callback Two new analysis sections have been added to post-call conversation analytics: * **DND (Do Not Disturb)** β€” Surfaces DND signals detected during the call with configurable expiration windows and timezone-aware logic. * **Callback** β€” Tracks callback intent and scheduling outcomes, including automatic campaign-based callback scheduling after analysis is generated. These sections give a richer picture of call outcomes beyond standard transcription. *** ### πŸŽ™οΈ Voice Engine β€” Cartesia & Deepgram TTS Streaming New streaming TTS providers are now available: * **Cartesia TTS Streaming** β€” Streaming audio generation using Cartesia's latest models, with built-in retry logic. * **Deepgram TTS Streaming** β€” Deepgram TTS now streams audio with full web-call integration, barge-in support, caching, and retries. *** ### πŸ—£οΈ Deepgram Multilingual STT β€” Flux Multi Model The **Deepgram Flux Multi** model is now available as a selectable speech-to-text option, bringing multilingual speech recognition to assistants that serve non-English speakers. *** ### πŸ‘€ User Management β€” Developer Role A new **Developer** role has been added to the role hierarchy, sitting between User and Admin. #### Role hierarchy ``` Admin β†’ Developer β†’ User ``` The Developer role is designed for engineering teams who need to build and operate platform resources without access to sensitive administrative controls. #### What developers can do | Module / Feature | Access | | ------------------------------------- | :----: | | Assistants | βœ… | | Workflows | βœ… | | Conversations & Outbound Calls | βœ… | | Campaigns | βœ… | | Integrations & Knowledge Bases | βœ… | | Call / Chat / Customer / Webhook Logs | βœ… | | Workflow Runs & Audit Logs | βœ… | #### What developers cannot do | Module / Feature | Access | | ------------------------------- | :----: | | Users Management | ❌ | | Phone Numbers Management | ❌ | | API Keys Management | ❌ | | Provider Credentials Management | ❌ | | Feature Settings | ❌ | | Organizations Management | ❌ | Admins retain full control over sensitive configurations β€” credentials, API keys, user management, and organization-level settings remain restricted to Admin and above. *** ### πŸ”’ Workflow & Assistant Lock Both workflows and assistants can now be **locked** to prevent concurrent edits while they are in active use. * Lock/unlock via a dedicated API (admin users only). * Full audit trail for lock state changes. *** ### πŸ› Bug Fixes & Enhancements * Fixed a race condition where concurrent requests could silently overwrite conversation update data. * Fixed the hold command (`playAndBargein`) not behaving correctly in certain playback states. * Fixed audio being incorrectly stopped on dialout after a hold or during background sound playback. * Fixed post-call transcript speaker labels being incorrectly assigned. * Added automatic retries and duplicate-prevention for failed post-call transcript processing. * Fixed ElevenLabs WebSocket streaming reliability including background sound handling. * Fixed Twilio bulk schedule call 500 errors. * Added strict Twilio webhook signature validation. * Upgraded aiohttp, cryptography, litellm, and libssl3/curl to address known CVEs. ## What's new in v1.13.0 πŸš€ Updates & Improvements ### πŸŽ™οΈ Voice Engine **Upgrades to TTS, realtime voice, and call performance** #### Dynamic TTS Speed Control You can now control TTS playback speed dynamically at runtime via commands. Speed values are validated before being applied, preventing out-of-range values from causing unexpected audio behavior. #### Deepgram TTS WebSocket Support Deepgram TTS is now available over WebSocket, including full webcall integration. This improves reliability and reduces latency for Deepgram-powered voice playback. #### Advanced Voice Configurations Support for `gpt-realtime-1.5` has been added, along with exposed advanced realtime voice settings that were previously inaccessible. These settings allow fine-grained control over realtime voice behavior for assistants using OpenAI's realtime API. #### Workflow Performance β€” Cache Pre-warming Workflow-based assistant performance has been improved by pre-warming the workflow cache in the background at call creation time. This reduces cold-start latency on the first workflow invocation during a call. #### Hold Command Bug Fix Fixed an issue with the `playAndBargein` hold command that could cause incorrect call handling behavior when a hold was triggered during active playback. *** ### πŸ“ž Call Handling & Integrations #### Twilio Inbound Callhook New APIs for retrieving and binding Twilio inbound callhooks are now available. Minor updates have also been made to number management to support inbound callhook configuration. #### Twilio Signature Validation Twilio request signature validation has been improved. Rejection of requests with invalid signatures has been temporarily disabled to ensure reliability during the rollout, while the validation logic itself is hardened. **Note**: Signature rejection will be re-enabled in a future release once validation is fully verified across all inbound traffic patterns. #### Active Calls Query Optimization The active calls query has been fully rewritten to avoid inefficient lookups. This improves performance for teams with a high volume of concurrent calls and reduces database load during peak traffic. *** ### πŸ“ Transcripts & Reporting #### Offline Call Transcripts Call transcripts can now be uploaded to S3 after a call ends. This release also adds: * A dedicated transcript download API * Improved Deepgram response parsing for more accurate transcript data * A new assistant-level config for enabling post-call transcript generation Use these features to archive call transcripts for compliance, quality review, or downstream analytics. #### Conversation Analysis Enhancements Two new analysis sections have been added to conversation analytics: * **DND (Do Not Disturb)** β€” surfaces DND signals detected during the call * **Callback** β€” tracks callback intent and scheduling outcomes These sections give a richer picture of call outcomes beyond standard transcription. #### Bug Fixes * **DND null check** β€” Fixed a null reference error that could occur when a DND object was absent during a DND update operation *** ### πŸ”„ Workflow Service **Significant runtime improvements, new model support, and observability upgrades** #### Anthropic Claude Model Support Anthropic Claude models are now supported within the workflow framework alongside existing OpenAI and Azure models. Select a Claude model in your LLM node configuration to use it in any workflow. #### Workflow Node Builders Pre-built node builders are now available for the following node types, reducing the boilerplate required to add common workflow steps: * `summary` β€” summarize conversation context * `success` β€” mark a workflow run as successful * `structured_data` β€” extract structured output from an LLM response * `dnd` β€” handle Do Not Disturb detection * `callback` β€” schedule a callback action #### Workflow Run Checkpointing Workflow runs now support checkpointing and state reconstruction. If a run is interrupted, execution can be reconstructed from the last checkpoint rather than restarting from the beginning. #### Resume Workflows from Any Node Workflows can now be resumed from any node with an initial state context provided at resume time. This enables retry and recovery flows without requiring a full restart. #### Workflow Cache Improvements The workflow cache is now enabled by default. An internal endpoint for manual cache management has also been added for operational use. #### Redis Caching for Workflows Redis caching for fully hydrated workflows has been implemented behind a feature flag (disabled by default). When enabled, this significantly reduces workflow load times for frequently-used workflows. #### In-Memory Workflow Handler An in-memory workflow handler has been added behind a feature flag. This is intended for environments where low-latency execution is prioritized and external state persistence is managed separately. #### Evaluation Workflow Run Persistence Evaluation workflow runs are now persisted to the workflow service, making it possible to review and compare evaluation results over time. #### Debug Mode Observability Debug mode now captures individual LLM metrics per node execution, giving developers per-node visibility into model latency and token usage during a workflow run. #### OpenAI Realtime Support Support for `gpt-realtime-1.5` has been added to the workflow service, along with dynamic Realtime API configuration. This aligns the workflow runtime with the voice engine's realtime capabilities. #### Fixes & Dependency Updates * Fixed timezone expectation handling in workflow execution * Fixed empty-messages edge case that could cause unexpected LLM behavior * Improved credential redaction in workflow logs to prevent accidental secret exposure * Updated dependencies for in-memory workflow runtime: `beanie`, `langchain`, `boto3` *** ### πŸ“£ Conversation & Campaigns #### Twilio Throughput Limit A Twilio throughput limit of 75 MPS has been implemented for sending campaign pre-call messages. This prevents campaigns from exceeding Twilio's rate limits and causing message delivery failures at scale. #### Athena Appointment Rescheduling Appointment rescheduling is now supported for Athena integrations. Assistants can handle rescheduling requests end-to-end through the EHR, in addition to the existing scheduling and cancellation flows. ## What's new in v1.12.0 πŸš€ Updates & Improvements ### πŸ“£ Campaigns #### Retry Mechanism Full retry policy support is now available for campaigns. This includes: * Retry configuration models * CRUD APIs for managing retry policies * Retry logic for individual calls and bulk campaign execution Configure how many times a failed call should be retried and under what conditions before it is marked as permanently failed. #### Contacts Management A new contacts management module has been added to the campaigns service with dedicated CRUD APIs. Contacts can be created, listed, updated, and deleted independently of campaigns β€” making it easier to manage reusable contact lists across multiple campaigns. #### Call Filtering & Statistics New API endpoints for filtering calls by outcome and retrieving campaign-level statistics are now available. Use these to build dashboards, slice data by outcome or status, and measure campaign performance. ## What's new in v1.11.0 πŸš€ Updates & Improvements ### πŸ” Workflow Enhancements **Major platform-level upgrades to the workflow engine** This release ships a large set of improvements to the workflow runtime, tooling, observability, and model support. #### Callback Scheduling Workflows can now schedule future callbacks via a dedicated API endpoint with full validation. Dynamic variable handling has been extended to support callback scheduling flows, so you can defer execution to a future time without custom workarounds. #### ToolNode β€” Direct Tool Execution A new `ToolNode` node type lets you invoke tools directly within a workflow without routing through an LLM. This is useful for deterministic steps where you want predictable, low-latency execution without incurring an LLM call. #### GPT-5 & GPT-5.4+ Model Support ```javascript theme={null} { "model": "gpt-5", // or "gpt-5.4" "reasoning_effort": "high" // configurable effort level } ``` GPT-5 and GPT-5.4+ are now supported with configurable `reasoning_effort` levels. Token usage is tracked uniformly across all LLM providers. #### OpenAI Responses API Native support for the OpenAI Responses API is now available with an explicit opt-in flag. The `reasoning_effort` parameter is passed through for custom LLM configurations. #### External API Tool Enhancements The external API calling tool has been significantly upgraded: * **Native response types preserved** β€” responses are no longer cast to string * **Explicit error propagation** β€” errors surface to the workflow instead of being silently swallowed * **HTTP retry support** β€” transient failures are automatically retried * **Dynamic argument substitution** β€” LLM tool arguments can be injected into the URL path, query parameters, and request body #### Liquid Template Expressions in Dynamic Variables ```liquid theme={null} {{ patient.first_name | capitalize }} {{ patient.last_name }} ``` Dynamic variables now support Liquid template expressions. Expressions are evaluated and cached during workflow execution, enabling rich templating without writing custom tool code. #### Advanced Dynamic Variables Extended dynamic variable operations with predefined system-level variables, giving workflows access to runtime context such as timestamps, call identifiers, and team metadata. #### Granular Error Tolerance Configure per-workflow and per-node error tolerance thresholds so that non-critical failures degrade gracefully instead of aborting the entire run. #### LLM Latency & Call Tracking Aggregated LLM latency statistics and call count are now tracked across workflow runs, giving you detailed observability into model performance over time. #### Vendor Credential Caching Vendor credentials are now cached using a Redis cache-aside pattern with secure fallback. Credentials are scoped per `team_id`, so multi-tenant deployments remain isolated. #### Workflow Runtime Stability Improved error handling prevents cascading failures when individual nodes encounter errors. #### Azure OpenAI Fix Resolved a validation error triggered by a missing `api_version` field in Azure OpenAI chat configurations. #### UI Ordering in System Prompt Added UI ordering metadata and a special field type to system prompt configuration for workflow nodes, making it easier to organize and read complex prompt structures in the dashboard. *** ### πŸ’¬ Conversations #### Start Message on Tool Execution When a non-interruptible music tool begins executing during a conversation, a start message now plays automatically. This provides callers with immediate audio feedback that the system is processing, reducing perceived silence. *** ### πŸ”‘ Vendor Credentials #### TTS Playground Vendor Credential Support The TTS playground now correctly uses vendor-level credentials when testing voices. This allows you to test provider-specific voices using your own credentials, rather than falling back to default platform credentials. #### Redis Cache Invalidation for Vendor Credentials Vendor credential updates are now immediately propagated across the system via Redis cache invalidation. Previously, credential changes could take time to take effect due to stale cache entries. *** ### πŸ› οΈ Admin Tools #### Webhook Logs Admins can now view webhook activity logs directly in the dashboard. This provides improved observability into webhook delivery, making it easier to debug integrations and verify that events are being received correctly. *** ### πŸ› Bug Fixes #### Assistant Call History Fixed persistence of assistant call history, including the conversation ID link and assistant phone number resolution. Call history entries were previously missing or incorrectly linked in some cases. ## What's new in v1.7.0 πŸš€ Updates & Improvements ### πŸ” Okta SSO Login Support **Enterprise Single Sign-On Integration** We've added support for Okta Single Sign-On (SSO), enabling seamless integration with enterprise identity providers. Okta SSO support is available for organizations that require enterprise authentication. **Note**: Okta SSO is not enabled by default. If you need SSO support for your organization, please contact our support team and we'll enable it for you. ### 🎡 Hold-Call Tag Support **Pause Calls with Custom Hold Music** Custom LLM workflows can now trigger call holds with configurable hold music using the hold-call tag. This is useful for scenarios where you need to pause a call and play background music. #### How to Use Add the following XML tag to your custom LLM response: ```xml theme={null} keyboard0.5 ``` #### Configuration Details * **``** - Triggers the hold state * **``** - Type of hold music (e.g., `keyboard`, `music`) * **``** - Duration in seconds to play hold audio When the hold-call tag is detected, the call will be paused and hold music will be played for the specified duration before the call resumes. ### πŸ” Search KB and Assistants by ID **Resource Discovery by Object ID** Our dashboard portal now supports searching for knowledge bases, assistants, and workflows by their object IDs, in addition to existing search by name and description. #### Features * **Search by ID** - Find resources directly using their unique object IDs * **Quick Navigation** - Click search results to instantly navigate to the resource detail page * **Comprehensive Search** - Combine ID-based search with name and description searches for maximum flexibility This enhancement makes it faster to locate resources, especially in organizations with many assistants and knowledge bases. ### πŸ”„ Redis Retry Handling for Connection Failures **Improved Infrastructure Resilience** We've enhanced our Redis connection handling to gracefully manage connection failures, particularly during cloud provider maintenance windows or temporary service interruptions with AWS Redis Cache and similar services. #### What's Improved * **Automatic Retry Logic** - Seamless recovery from temporary Redis connection failures * **Better Error Handling** - Improved internal error management to prevent cascade failures * **Enhanced Stability** - More reliable system operation during infrastructure maintenance * **Transparent Recovery** - Automatic reconnection without manual intervention **Impact**: This improvement reduces system errors and improves overall platform reliability during cloud provider maintenance windows and temporary outages. ### πŸ“¨ Webhook for Concurrency-Failed Call Events **Improved Event Delivery for Failed Calls** We've fixed webhook delivery for calls that exceed your organization's concurrency limits. Previously, `end-of-call-report` and `status-update` webhooks were not delivered when calls hit concurrency limits. #### What's Fixed * **Concurrency Limit Events** - Webhooks are now delivered when calls fail due to concurrency limits * **Consistent Event Delivery** - Both `end-of-call-report` and status-update events with failed status are now reliably delivered * **Assistant-Level Configuration** - Webhooks are sent to configured webhook servers at the assistant level **Pro Tip**: Monitor concurrency failure webhooks to track and optimize your concurrency limits. ### βœ… Campaign Pre-Call Validations **Validate Calls Before Campaign Launch** A new field called "Validate Call Data Before Launch" has been added to campaign creation and update workflows. This enables you to validate all calls before your campaign goes live. #### How to Use Enable call validation in your campaign settings: ```javascript theme={null} { "validateOnUpload": true } ``` #### Benefits * **Quality Assurance** - Catch invalid data before campaign execution * **Early Error Detection** - Identify issues with phone numbers, formatting, and data integrity * **Webhook Notifications** - Failed validations are reported to your configured webhook server * **Available Everywhere** - Supported via both API and dashboard UI When enabled, all calls in your campaign are validated before the campaign launch time. Any failed validations will trigger webhook notifications to your configured endpoint, allowing you to address issues proactively. **Recommendation**: Enable this feature to ensure campaign data quality and prevent execution delays. ### ⏰ Campaign Call Schedule Timing Overrides **Flexible Per-Call Schedule Configuration** Campaign calls now support schedule timing overrides, allowing individual calls within a campaign to use different timezones and execution times than the overall campaign schedule. #### How to Use Set override parameters per call during campaign upload: ```javascript theme={null} { "override_timezone": "America/New_York", "override_schedule": [ { "startAt": "2025-09-27T12:00:00.000Z", "endAt": "2025-09-27T20:00:00.000Z" } ] } ``` #### Example Scenario Suppose your campaign is scheduled to run today and tomorrow from 9 AM to 6 PM: * **Overall Schedule**: Today & Tomorrow, 9 AM - 6 PM * **Override Example**: Some calls run only today 3 PM - 5 PM * **Override Example**: Other calls run only tomorrow 10 AM - 11 AM #### Configuration Methods * **API** - Using `override_timezone` and `override_schedule` fields * **Dashboard UI** - Campaign calls upload screen This gives each call in your campaign maximum flexibility for scheduling. ### πŸŽ™οΈ Campaign Assistants Overrides **Per-Call Assistant Configuration** Campaign calls can now override specific assistant features, allowing you to customize assistant behavior (voice, welcome message, and other configurations) for individual calls within a campaign. #### Override Assistant Configuration Partially override assistant settings for a specific call: ```javascript theme={null} { "override_assistant": { "voice": "custom-voice-id", "welcomeMessage": "Custom greeting here", // ... other assistant configuration fields } } ``` #### Override Assistant Entirely Completely replace the campaign assistant with a different permanent assistant: ```javascript theme={null} { "override_assistant_id": "assistant-uuid-here" } ``` #### Priority Rules * **Both Fields Provided** - `override_assistant_id` takes priority over `override_assistant` * **`override_assistant_id` Only** - Use a completely different assistant * **`override_assistant` Only** - Customize specific fields of the campaign assistant #### Configuration * **API** - Supported via API only * **UI** - Coming in future releases This feature enables fine-grained control over assistant behavior on a per-call basis within campaigns. ### πŸ“Š Campaign Calls Metadata Tracing **Custom Metadata for Call Tracking** Campaign calls now support custom metadata that you can use to track and correlate calls within your systems. Metadata can be configured during campaign call upload and is returned in webhook events. #### How to Use Configure metadata during campaign call upload: ```javascript theme={null} { "override_metadata": { "customerId": "customer-123", "campaignPhase": "phase-2", "externalRef": "external-id-456" } } ``` #### Features * **Custom Tracking** - Define any metadata key-value pairs for your use case * **Webhook Integration** - Metadata is returned in all call-related webhooks * **Call Correlation** - Use metadata to correlate calls with your internal systems * **API Only** - Supported via API (`override_metadata` field) The metadata object values you provide will be passed to the call's metadata and returned in all subscribed webhook events, making it easy to track and correlate calls within your own systems. ### πŸ› Bug Fixes and Improvements * Enhanced system stability and reliability across platform * Improved error handling for edge cases * Better performance optimization for large-scale operations * UI/UX refinements for improved user experience * Fixed various minor issues affecting call quality ## What's new in v1.6.0 πŸš€ Updates & Improvements ### πŸ“ž Campaign Call Validation Before Launch **Validate Calls Before Campaign Execution** We've introduced a new feature that allows you to validate campaign calls before the campaign is launched. Previously, validation occurred only after the campaign had been launched, but now you have the ability to validate calls beforehand, ensuring better quality control and error prevention. #### How to Use Enable call validation via API by setting the new `validateOnUpload` field: ```javascript theme={null} { "validateOnUpload": true // Enable pre-launch call validation } ``` #### Configuration Details * **Field Name:** `validateOnUpload` * **Type:** `Boolean` * **Default:** `false` * **Available via:** API **Pro Tip**: Enabling `validateOnUpload` helps catch invalid phone numbers, formatting issues, and other potential problems before your campaign goes live, saving time and improving campaign success rates. ### πŸŽ™οΈ Transcriber (STT) Fallback Support **Enhanced Speech-to-Text Reliability** We've added fallback support for transcribers (Speech-to-Text). You can now enable fallback transcribers to ensure continuous speech recognition even if the primary transcriber encounters issues. This enhancement provides: * **Improved Reliability** - Automatic failover to backup transcriber * **Uninterrupted Conversations** - Seamless transition during transcription issues * **Better Call Quality** - Reduced chances of missed or garbled transcriptions ### πŸ”Š Self-Hosted Deepgram Transcriber Support **New Self-Hosted STT Option** We now support self-hosted Deepgram as a transcriber (STT) option, giving you more control over your speech-to-text infrastructure. #### Supported Models Currently supported models: * **Nova 2** - English language * **Nova 3** - English language **Coming Soon**: We will be adding support for more models and additional languages in subsequent releases. ### πŸ€– Gemini 3-Flash LLM Support **New LLM Model Option** We've added support for **Gemini 3-Flash** as an LLM model option for your assistants. This new model provides: * Fast response generation * High-quality conversational capabilities * Cost-effective performance for real-time interactions You can now select Gemini 3-Flash from the LLM model options when configuring your assistant. ### πŸ› οΈ Bug Fixes and Enhancements * Resolved various issues affecting system stability and performance * General improvements to call quality and user experience * Enhanced error handling across the platform * UI/UX refinements for better usability ## What's new in v1.3.0 πŸš€ Updates & Improvements ### πŸ”§ External API Tools - Full HTTP Method Support **Enhanced API Integration Capabilities** External API Tools now support all standard HTTP methods, providing complete flexibility for integrating with third-party services and APIs. #### What's Changed **Before:** Only `POST` method was supported by default for External API Tools. **Now:** Full support for all HTTP methods: * `GET` - Retrieve data from APIs * `POST` - Create new resources * `PUT` - Update existing resources completely * `PATCH` - Partial updates to existing resources * `DELETE` - Remove resources #### How to Use 1. Navigate to the **Tools Section** in your dashboard 2. Create or edit an External API Tool 3. Select your desired HTTP method from the dropdown 4. Configure the tool parameters as needed 5. Attach the tool to your assistant This enhancement enables more sophisticated integrations with REST APIs, allowing your assistants to perform comprehensive CRUD operations and interact with a wider range of external services. ### πŸ“š Knowledge Base Optimization **Improved Query Performance and Indexing** * **Enhanced Query Processing** - Optimized algorithms for faster knowledge base searches and more accurate results * **Advanced Indexing** - Improved indexing mechanisms that reduce response times and enhance search relevance * **Better Scalability** - Enhanced performance for larger knowledge bases with more documents and data These improvements result in faster knowledge retrieval during conversations and more accurate responses from your assistants. ### 🎀 ElevenLabs Configuration Enhancement **New Streaming Latency Control** We've added granular control over ElevenLabs streaming optimization, giving you more precise control over the balance between quality and latency. #### New Configuration Field ```javascript theme={null} // New configurable field in ElevenLabs TTS settings optimizeStreamingLatency: Number // Range: 0 (max quality) to 4 (max speed) ``` #### Configuration Details * **Field Name:** `optimizeStreamingLatency` * **Value Range:** `0` to `4` * **New Default:** `0` (previously managed internally with value `3`) * **Configuration Methods:** * Dashboard UI toggle * API configuration support #### Latency Optimization Levels * **`0`** - Maximum quality, higher latency (recommended for most use cases) * **`1-2`** - Balanced quality and latency * **`3`** - Previous default - good balance * **`4`** - Maximum speed, optimized for real-time interactions **Migration Note**: Existing configurations will automatically use the new default value of `0`. If you were relying on the previous behavior, you may want to set this to `3` to maintain the same performance characteristics. ### πŸ“Š Per-Turn Latency Metrics **Enhanced Performance Monitoring** We've introduced detailed per-turn latency metrics for both user and assistant interactions, providing unprecedented visibility into conversation performance. #### New Metrics Available **User Turn Metrics:** * Audio Packets latency * Voice Activity Detection (VAD) latency * Speech-to-Text (STT) processing time * Total user turn processing duration **Assistant Turn Metrics:** * Large Language Model (LLM) response generation time * Text-to-Speech (TTS) processing duration * Total assistant response latency #### Where to Find These Metrics 1. **Debug Mode Dashboard** * Enable Debug mode from your profile or by clicking the Interactly.ai logo * Click on individual user or assistant messages in Call Logs * View detailed timing breakdowns for each interaction 2. **Webhook Integration** * Available in webhook subscription `message` events * Real-time access to latency data for your applications * Perfect for monitoring and analytics systems #### Use Cases * **Performance Optimization** - Identify bottlenecks in conversation flow * **Quality Monitoring** - Track response times across different configurations * **Analytics Integration** - Export timing data to your monitoring systems * **Debugging** - Diagnose latency issues in real-time conversations ### πŸ› οΈ Bug Fixes and Improvements * Resolved minor issues affecting call stability and user experience * Various UI/UX refinements for better user interaction ## What's new in v1.2.0 πŸš€ Updates & Improvements ### πŸ”§ Stability Improvements * Enhanced system reliability with comprehensive bug fixes and performance optimizations * Resolved known issues that were impacting call quality and user experience * Improved error handling and recovery mechanisms across the platform ### 🎨 UI Enhancements * Refined user interface elements for better visual consistency and usability * Updated component styling and layouts for improved user experience * Enhanced accessibility features and responsive design improvements ### πŸŽ™οΈ Noise Suppression Configuration Updates **⚠️ Breaking Changes - Assistant Configuration** We've improved the noise suppression configuration with a more comprehensive and flexible approach. The following fields are now **removed**: ```javascript theme={null} // ❌ Removed - These fields are removed completely and no longer supported noiseSuppressor: Boolean noiseSuppressionKnob: Number ``` **New Configuration Structure** Replace the removed fields with the new `noiseSuppressorConfig` object: ```javascript theme={null} // βœ… NEW - Enhanced noise suppression configuration noiseSuppressorConfig: { enabled: { type: Boolean, default: false }, suppressBackgroundNoise: { type: Boolean, default: true }, // MVNS flag suppressBackgroundVoice: { type: Boolean, default: false }, // BVS flag digitalGainControl: { type: Boolean, default: true }, // DGC flag knobValue: { type: Number, default: 100 } } ``` #### Configuration Options Explained * **`enabled`** - Master toggle for noise suppression functionality * **`suppressBackgroundNoise`** - Controls Multi-Variate Noise Suppression (MVNS) to filter out ambient sounds * **`suppressBackgroundVoice`** - Controls Background Voice Suppression (BVS) to minimize interference from other speakers * **`digitalGainControl`** - Controls Digital Gain Control (DGC) for automatic volume adjustment * **`knobValue`** - Fine-tune suppression intensity from 0 (minimal) to 100 (maximum) #### Migration Guide **Before:** ```javascript theme={null} { noiseSuppressor: true, noiseSuppressionKnob: 80 } ``` **After:** ```javascript theme={null} { noiseSuppressorConfig: { enabled: true, suppressBackgroundNoise: true, suppressBackgroundVoice: false, digitalGainControl: true, knobValue: 80 } } ``` **Migration Required**: Please update your assistant configurations to use the new `noiseSuppressorConfig` object. The deprecated fields will continue to work temporarily but will be removed in a future release. ### πŸ› οΈ General * Performance improvements and system optimizations * Enhanced logging and monitoring capabilities * Security updates and compliance improvements ## What's new in v1.1.0 πŸš€ Updates & Improvements ### πŸ“ž Caller ID Fix for Outbound Forwarded Calls When an assistant‐initiated outbound call is forwarded to another recipient, the user caller ID (user’s phone number) is now preserved. The receiving user will no longer see the forwarded assistant’s caller ID. ### 🎨 UI Enhancements * **Call Logs:** Error Commands/Utterances are now visually highlighted in **red** for quick identification. * **Customer Logs:** Updated layout and improved readability for streamlined troubleshooting. * **Assistant STT Configuration:** UI support added for **Deepgram Flux** model configuration. ### πŸ—£οΈ Assistant Behavior Enhancements * Added support for `` tag to ensure uninterrupted prompt/audio playback by the assistant. * Webhook listeners now emit **LLM error events**, allowing better error tracking and observability. ### πŸ“š Knowledge Base * Pagination and search capabilities added within the assistant’s Knowledge Base for faster navigation and scaling with larger datasets. ### πŸ”„ Integration & Automation * Support added to run **cadence flow using Microsoft integration**. ### πŸ“˜ API Docs Optimizations * Improved API documentation layout and navigation for easier reference. * Certain API URLs have been updated to new paths. If you have bookmarked API endpoints, please review and update them accordingly. ### πŸ› οΈ General * Multiple bug fixes and performance improvements across the platform. ## What's new in v1.0.0 ### 1. Enable Debug mode (see **Enable Debug mode** section below) Quickly turn on verbose developer logs from your profile so you can inspect detailed runtime information when diagnosing issues. When Debug mode is enabled the dashboard surfaces expanded logs and error traces useful for developers during troubleshooting. See the [**Enable Debug mode**](#enable-debug-mode) section below for more details. ### 2. TTS β€” new audio cache scopes (assistant & team) + clip deletion We added two new cache scope levels for TTS audio clips β€” **assistant** and **team** β€” so generated audio can be cached at the most appropriate scope for reuse and cost savings. The dashboard now also exposes the ability to delete existing audio clips so you can manage storage and refresh voices or content when needed. ### 3. Campaign Webhooks You can now register webhooks to receive real-time notifications for important campaign lifecycle events (for example: campaign completion) and per-call status updates (for example: call completed, failed, or dropped). Webhooks may be configured when **creating** or **updating** a campaign, enabling easy integration with downstream systems and automation pipelines. ### 4. Twilio SMS β€” Inbound & Outbound Dashboard toggle to enable Twilio SMS inbound and outbound functionality. This makes two-way SMS possible so your flows can receive replies and send messages via Twilio directly from the dashboard β€” useful for bi-directional support and conversational workflows. ### 5. Vonage Number management Buy or import Vonage phone numbers directly from the dashboard. Note: you will need to provide your Vonage credentials when adding or importing numbers. ### 6. Fixed: missed tool calls and trailing messages Resolved a bug where tool calls and messages at the end of conversations could be missed or dropped. This fix improves reliability for trailing-message flows and integrations that depend on final tool outputs. ### 7. Stability & performance improvements General reliability and performance upgrades across the platform β€” faster page loads, reduced error rates, and smoother dashboard interactions. ## Enable Debug mode Debug mode helps developers view detailed diagnostic logs and metadata for conversations, making it easier to analyze performance, latency, and behavior of both Assistant and User utterances. *** ### How to enable Debug mode You can enable Debug mode in two ways: 1. **From the logo** β€” Click the **Interactly.ai** logo at the top-left corner. Click On Interactly.ai Logo 2. **From your profile** β€” Click the **Profile** icon at the top-right and toggle **Debug mode** ON. Click On Profile Icon Once enabled, additional debug details will appear inside the **Call Logs** and **Conversation view**. *** ### Debug info in Call Logs Open any conversation from **Call Logs** after enabling Debug mode. You'll notice new metrics displayed at each **Assistant** and **User** utterance. Click on Agent transcripts #### Assistant utterance metrics * **AI: 1093 ms** β€” Time taken by the LLM to generate the response. * **TTS: 356 ms** β€” Time taken by the TTS engine to generate the audio clip. * **AD** β€” Audio Duration of that particular clip. #### User utterance metrics * **AD** β€” Audio Duration of the user's spoken input. * **VAD: 6 ms (Voice Activity Detection)** β€” Time the STT vendor waited after receiving the final word. * **AVAD: 800 ms (Additional Voice Activity Detection)** β€” Extra waiting time configured at the Assistant level under **Advanced β†’ Start Speaking Plan β†’ Smart Endpointing OFF**. Click On User utterances *** ### Accessing Meta Info β€” Assistant utterances To view detailed metadata, click on the **Assistant** key in the utterance view. Accessing Meta Info #### Key fields in meta info * **command** β€” Indicates whether the action is `Play` or `End`. * `End` signals the call termination. * **messageId** β€” Unique ID for the Assistant's utterance. * **userMessageId** β€” The ID of the corresponding user message that triggered this response. #### AI Meta fields * **finish\_reason** β€” Can be `stop`, `tool_calls`, or `length`. * `stop`: Response finished naturally. * `length`: Token limit reached. * `tool_calls`: Model triggered a tool/function. * **queue\_latency** β€” Time the request waited before model processing started. * **response\_latency** β€” Time the model took to generate the response. * **trailing\_messages** β€” Last 6 utterances passed to the LLM for context. #### Other useful fields * **audioDuration** β€” Duration of the generated audio clip. * **isBargein** β€” Whether the utterance was interrupted. * **isCachePlaying** β€” `true/false`; indicates if the clip was played from cache or generated fresh. * **model / vendor** β€” The TTS model and vendor used. *** ### Accessing Meta Info β€” User utterances Click on the **User** key in the utterance to view metadata for the user's input. Click On User Key #### Important fields * **messageId** β€” ID of the user message. * **previousBotMessageId** β€” ID of the last Assistant response before this message. * **skippedBotMessages** β€” List of interrupted Assistant messages skipped due to this utterance. * **confidence** β€” Confidence score returned by the STT engine for this transcript.