Since November 12, on Enovacom GXD5 centers with a very large configuration,
refreshing the copy of the center's exams, sites, rooms and practitioners could
restart the booking service, about once an hour of calls. Booking requests in
progress at that moment failed, on Enovacom and on other booking software alike.
The copy now holds only what the assistant uses, and storing or reading it no
longer takes enough memory to restart the service.
The Reecall MCP server supplies dashboard links so connected AI agents can direct you to your Assistants and other resources.
Bug fixes
Requests to repeat no longer talk over the caller
When an Assistant cannot transcribe what a caller said while it was speaking, it waits 1.5 seconds after it finishes before it asks the caller to repeat. If the caller speaks during that time, the Assistant does not ask.
The caller can interrupt the request to repeat, even when disableInterruptions is true. The Assistant does not ask again until it transcribes the caller's speech. Before, with disableInterruptions, the request could repeat several times in a row.
When a caller books several exams, the exam order set in the booking
configuration now also holds across days. An exam set to pass first is never
booked on a later day than the exam that follows it. Before, the order only
applied to exams on the same day. This covers the order per exam category and
the order rules per exam.
If the order rules form a loop, such as A before B, B before C and C before A,
no booking can respect them. The assistant then offers no slot and transfers the
call.
Added send_email so your Assistant can email configured recipients during a call, such as notifying the front desk that a caller will be late. See Email tool.
Improvements
Faster first reply on calls that use a Mistral model.
Post-call data extraction retries stalled requests, reducing calls without extracted data.
Bug fixes
resolve_datetime counts days_from_now in calendar days in the Assistant's time zone. Tomorrow is 1, even when it is less than 24 hours away.
When a transfer destination is missing, the Assistant picks one of your configured destinations before announcing the transfer. This prevents an announcement followed by a failure.
Data API
Improvements to Mapper feature
The date helper reads day-first dates (01/03/2024 is March 1) and Excel serial numbers. A date it cannot read now fails the row instead of being read month first. For month-first files, use the new parse_date helper: {{ Date | parse_date('MM/dd/yyyy') | date('yyyy-MM-dd') }}
POST /automation/mappers/{id}/run returns a 400 with BAD_INPUT for a malformed body, instead of a 500.
Saving a mapper validates inputOptions.
A JSON source made of one object that wraps the rows, such as { "items": [...] }, is rejected instead of being read as one row.
Blueprints require a type: assistant, mapper, or automation.
The type must match the models the Blueprint creates. A Blueprint whose models don't match its type is rejected.
Existing Blueprints have been given the type that matches their models.
If you create Blueprints through the API, add type to your requests.
Security
Blueprint access between projects
Fixed access to Blueprints from other projects through a project-scoped API key. Each key only sees Blueprints from its own project.
AI Assistants
New features
Transfer a call to another Assistant
Added transfer_to_assistant so your Assistant can hand a conversation to another Assistant in the same organization and project without ending the call. See Assistant transfer tool.
Date and time resolution tool
Added resolve_datetime so an Assistant can look up a date's day of the week and distance from the current date during a conversation. See Date and time tool.
Fixes
Transfer summaries in the Assistant's language
The conversationSummary sent with a WebRTC transfer is written in the Assistant's language.
Added send_message so your Assistant can send an SMS to the caller or another number during a conversation.
Configure each tool with a message template, sender name, and the details the Assistant must collect to fill it in, such as a date, time, or reference.
Use variables and injected data in the text, such as the caller's name.
The collected details are checked against your rules before the SMS is sent.
On WebRTC calls or when the caller's number is hidden, the Assistant asks which number to send the SMS to.
GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna from OpenAI are available.
Claude Haiku 4.5 from Anthropic is available.
Gemini 3.8 Flash and Gemini 3.5 Flash-Lite from Google are available.
Speech-to-speech models
GPT-Realtime 2.1 and GPT-Realtime 2.1 Mini from OpenAI are available, with stronger tool calling and more reliable interruption handling.
Gemini 3.8 Live from Google is available as the stable successor to Gemini 3.1 Flash Live.
Speech-to-text models
Universal-3.5 Pro Realtime from AssemblyAI is available and recommended.
Text-to-speech models
Cartesia Sonic-3.5 is available as the drop-in successor to Sonic-3.
Improvements
Speech speed in Advanced Settings
Set tts.speed to control Assistant speech speed from -1, the slowest, to 1, the fastest. 0 or null keeps the integration's normal pace. See TTS settings.
Noise cancellation in Advanced Settings
Set audio.noiseCancellation to choose the caller's audio filter: bvc_telephony, the default, bvc, nc, or disabled. It applies to both LLM + TTS + STT and STS Assistants. See Audio settings.
Model lifecycle metadata
Model catalogue entries expose deprecation and discontinuation in their metadata, each with a date and, when available, the fallback model ID to migrate to.
The portal prevents selecting models whose discontinuation date has passed.
Deprecations and discontinuations
The following models are flagged as deprecated or discontinued by their provider.
When a caller is inactive, the Assistant prompts them three times at 5-second intervals before ending the conversation. If the caller responds, the conversation resumes.
Fixed a bug where the Assistant waited indefinitely for the WebRTC client to resolve a transfer request. The Assistant ends the conversation after 5 minutes without a resolution.
Inject data when a conversation starts. Each key becomes a dynamic variable, resolved with {key} in the Assistant's instructions, first message, and other templated fields. See Data injection.
Configure dataExtractionSchema to extract data, such as caller intent, after a conversation ends. The result is stored in the Exchange's data property. See Data extraction.
Bug fixes
STS + TTS hybrid mode
Fixed a bug where Assistants using STS + TTS hybrid mode stayed silent.
Abrupt final sentence cut with STS
Fixed a bug where the final sentence from an STS Assistant was cut off before playback finished.
Data API
Features
Blueprints
Provision one or more resources in a single operation using a Blueprint. Each Blueprint defines the entities to create, with shared field injection through overrides and runtime value substitution through context.
Use named entries to reference resources created earlier in the same Blueprint, such as linking an Assistant to a Knowledge Base created in the same run.
Run a Blueprint inline via POST /core/blueprints/run, or store it as a reusable record and invoke it by ID via POST /core/blueprints/:id/run.
Project level external ID
Set the optional externalId field when creating or updating a Project with POST /core/projects or PATCH /core/projects/:id. Use it to store the identifier of an entity in your own system.
Bug fixes
Channel validation conflict
Fixed an API validation regression where SIP validation rules prevented WebRTC Channels from being created or updated.
Assistants can use STS instead of STT + LLM + TTS to reduce turn latency.
Knowledge Base support
Assistants can use your Knowledge Bases during conversations.
External ID support
Added externalId to associate conversations with identifiers from your external systems.
Added support for the custom SIP header X-External-ID.
Improvements
Exchange lifecycle management
Exchanges are created at conversation start with partial data and an inProgress status. At conversation end, they receive the complete conversation data and a completed status.
Gladia STT endpointing
Improved Gladia speech-to-text endpointing configuration for more accurate speech detection.
Bug fixes
Variable resolution
Fixed dynamic variable substitution in Assistant instructions.
Final Assistant sentence
Fixed a bug where the system hung up before the Assistant's final sentence finished playing.
Data API
Features
Speech-to-speech (STS) support (beta)
Added the STS entity for managing real-time voice interactions.
Added /ai/sts for STS configuration management, initially supporting gpt-realtime.
Knowledge Base support
Added /ai/knowledges to create knowledge entries.
Added /ai/knowledgeBases to group knowledge entries into Knowledge Bases.
Improvements
Enhanced API query capabilities
Added the select parameter to getById operations to choose response fields.
External ID support
Added externalId to the Exchange entity to associate it with an external system's identifier.
SIP Channels configuration enhancements
The /conversational/channels endpoint accepts krispEnabled and includeSipHeaders.
Exchange root data
Added conversation fields to the Exchange entity root:
duration: exchange duration in seconds
assistantId: ID of the Assistant that participated in the Exchange
Calls API
Features
External ID support
POST /calls/:channelId accepts externalId in its request payload.