Transcripts show the page a chat started on
Each chat transcript now records the page of your site where the visitor started the conversation, as the built-in vf-initial-url property. You find it in Transcripts > your transcript > Metadata.
Resumed sessions keep the first page, and values over 2048 bytes are cut short.
The property is recorded when the chat client sends the page it is running on in config.pageURL of the socket start.
Manage personas through the API, MCP and Atlas
You can now list, create, read, update and delete personas, the saved variable values a test or Run panel conversation starts from, through the stable API, the Voiceflow MCP server and Atlas.
- The stable API has persona endpoints scoped to a project and environment.
- The Voiceflow MCP server has a
voiceflow_personatool. - Atlas reads personas freely and treats create, update and delete as write actions.
- The OAuth consent screen offers a Personas permission group.
Each persona has a unique name and items that set an existing variable, by name, to a value; an update replaces the whole item list. Unknown or repeated variable names, duplicate persona names and going over the plan’s persona limit are refused with an explanation, and nothing is written.
Voiceflow MCP server now serves version 3 by default
The Voiceflow MCP server now gives AI clients one small vf_* tool per operation, so the tool list is smaller and cheaper. A client starts with vf_get_started, loads full schemas with vf_describe_operations, and loads how-to guides only when needed with vf_skill_list and vf_skill_load.
Add options to the https://mcp.voiceflow.com/mcp URL to narrow a connection:
toolsets=agent,playbooklimits the client to those tool groups.read_only=truegives the client read-only access.project=<projectID>limits the client to one project.
Clients that depend on the grouped voiceflow_* tools keep them by adding ?version=2 to the URL.
Attach metadata to a transcript property value
Integrations that tag transcripts with property values can now store structured context, such as who reviewed the transcript or a score, alongside each value. The Set property value endpoint of the stable API accepts it as an optional metadata object.
- Send
PUT /v1/stable/transcript-property/{propertyID}/valuewithvalue,transcriptIDand, if you want to record context,metadata. - Omit
metadataor passnullto record none.
Test a function against up to 12 cases via API
You can now run a function’s saved code against up to 12 test cases in one call, with no network access, before you rely on it. The Check function endpoint reports how each case compared with the expected path and outputs.
- Send a
POSTto the function check route with up to 12 cases, each withinputsand optionalexpectPathandexpectOutputs. - For each case, the response returns the exit path, the outputs, any error, and whether the case matched.
Check what a publish would change from the API
You can now compare an environment’s draft with its last publish from a script, the CLI or an AI client. It is the same comparison the publish modal shows, without opening the modal.
- Send a
GETto the environment diff route with aprojectIDand the environment alias. - The response returns the changes between the draft and the last publish, sorted into categories.
An environment that has never been published answers 409 and asks you to publish once first.
Search and set agent voices from Atlas, MCP or CLI
Atlas, the Voiceflow MCP server and the CLI can now find a voice in the ElevenLabs and Cartesia catalog and set it on your agent. Ask for “use Vikram” or “a calm British woman for French callers” and the voice is found and applied.
- Search voices (
GET /v1/stable/agent/voice-catalog) searches by words, voice ID, language, provider and gender, and returns a sample link for each voice. - Patch voices (
PATCH /v1/stable/agent/voices) sets the default voice, or adds, replaces or removes one language voice, without touching the agent’s other voices.
Voice IDs that are not in the catalog are refused.
Connect ChatGPT to the Voiceflow MCP server
ChatGPT can now be connected to your Voiceflow account through the Voiceflow MCP server, so you can work on your agents from ChatGPT.
- In ChatGPT, add a custom MCP server pointed at
https://mcp.voiceflow.com/mcpwith OAuth. - ChatGPT registers itself.
- Sign in to Voiceflow in the browser, and you are sent back to ChatGPT.
The connection signs in through MCP authentication.
Pick from full ElevenLabs and Cartesia voice catalogs
The voice pickers in your agent’s behaviour settings now load the complete ElevenLabs and Cartesia voice catalogs.
- Every vendor voice missing from the built-in list is added to the voices editor, alongside the other providers’ voices.
- Built-in voices show the catalog’s languages, accents, ages and use cases.
To browse them, open Settings > Behaviour > Voice and click a voice.
Translate transcripts in the transcript viewer
You can now translate a conversation into your own language from the transcript viewer, so you can read a conversation held in another language without copying it into a translator.
- Choose Translate in the transcript viewer under Transcripts to translate user messages, agent messages, card text and button labels.
- Choose Show original to switch back to the text as it was written.
- Markdown, links and numbers are kept in the translated text.
Each translation spends your workspace’s credits, and it is refused when they run out.
Queue messages to Atlas while it works
You can keep typing to Atlas while it works. Each message you send joins a queue above the input and goes out on its own when the turn finishes.
- A stacked deck reading “N queued. Sends when Atlas finishes.” appears when you press Enter or the send arrow while Atlas is working.
- Hover the deck to unfurl it, then choose Edit to put a message back in the input or Remove to delete it.
- Press the up arrow in an empty input to take back the newest queued message.
- Press Stop or Escape to return every queued message to the input and send none of them.
- The deck reads “Sends after you answer.” when Atlas is waiting on a question.
You can queue up to 10 messages, and plan sessions have no queue.
Analytics charts handle sparse and partial days
Charts in Analytics, Evaluations and Usage now join days without data with a faint dashed line and mark today’s unfinished period, so low-traffic agents no longer see lone dots or a false drop.
Pick voices from a full-screen voice library
You can now browse, preview and compare voices from different providers in one full-screen voice library, and pick a voice without choosing a provider first. To open it, click the Voice field under Voice output in your agent’s behaviour settings.
- Search voices by name or ID.
- Filter by language and accent, provider, model and gender.
- Play a sample before you click a voice.
- Picking a voice from another provider switches the provider for you.
- Choose the model in the Voice model field under the voice.
- Add ElevenLabs custom voices by ID or by pasting an ElevenLabs voice link.
Set a different voice for each language in Behaviour
You can now give your agent a different voice for each language it speaks, and manage all of its voices in one full-screen editor.
- Under Settings > Behaviour > Voice output in Behaviour, you see the primary voice and each language override with its flag, voice and provider, and Add language override adds a new one.
- Clicking a row opens the voices editor, with your agent’s voices on the left, the voice library in the middle and the selected voice’s model and settings on the right.
- In the left column you drag voices to reorder them and play them to preview, and your changes save as you go.
- The voice pill in the canvas header and the Agent voice link on the widget page open the same editor.
Connect your own SIP trunk to a phone number
You can now connect your own carrier’s SIP trunk, such as Twilio Elastic SIP, from Phone numbers > Add > Bring your own SIP. Inbound and outbound settings are separate, with failover addresses and encrypted call audio.
- Give the trunk a label, then turn on Inbound calls (allowed IPs, credentials or both) and/or Outbound calls and transfers.
- For outbound, add addresses in failover order, each with a port and a UDP, TCP or TLS protocol, plus an optional username, password and caller ID.
- Voiceflow gives you a SIP address to send calls to, and the trunk opens in a side editor that saves when you leave a field.
The optional Encrypted audio (SRTP) section is on both SIP forms and requires TLS.
Merge review warns before replacing Main’s unpublished draft
The merge review in Settings > Environments > Merge now warns you when the target environment has an unpublished draft that the merge will replace with the source’s published version.
Review Atlas prompt and instructions changes before approving
Before you approve an Atlas change to your global prompt, your agent’s instructions or a playbook’s instructions, you can see exactly what changes, with the stored text set against Atlas’s proposal.
- Click Review changes on the approval strip above the Atlas composer.
- Read the stored text against Atlas’s proposal in the changes review.
- Click Approve or Reject on the strip.
This applies when Atlas is in Manual mode.
Read and copy chunk metadata in the knowledge base
Hovering the grey metadata line above a chunk in a data source now shows a Metadata readout, and a Copy as JSON button in the tooltip copies it in one click.
- Each metadata key and its value get their own row, up to 10 rows, followed by an
and N morecount. - The smart-chunking options applied to the chunk are listed below the rows.
Find it under Knowledge base > Data source > Chunks.
Messages sent mid-turn interrupt your agent when possible
A message that arrives while your agent is busy, for example during a tool call, is now held on v4 socket and voice conversations instead of being rejected. It interrupts the turn at the first point where the turn can be interrupted, or runs when the turn ends.
- Several messages sent while one is waiting are merged into a single turn.
- A newer non-message action replaces an older held one, and the older one is rejected as superseded.
- A held end is never replaced.
Read evaluation release cards in full over the criteria table
A release card now opens in full over the criteria table when you hover a release marker on an evaluation card on the Evaluations page.
Close an evaluation’s log by clicking its row again
You can now close an Evaluation log by clicking its open row again in the Results table of an evaluation.
Clicking a different row still switches the log to that row.
Agents can link knowledge base source pages
Your agent can now link the page an answer came from: each result it reads from the knowledge base includes that page’s address, such as a web page URL or a Zendesk article’s public help center link. This holds even when Show source URL(s) is off.
- To have the agent link the page it used, ask it to in its instructions.
- To keep it from sharing links, tell it not to in the same place.
- Show source URL(s) still adds its separate citations in chat projects.
This works for agents that have the knowledge base tool turned on.
Shopify imports now include only active products
Importing a Shopify store as a data source now adds only active products to your knowledge base and the Add data sources count, not draft or archived ones.
Run panel debug logs show system prompt variables filled in
The ai result system prompt in Run panel debug logs now shows your variables filled in, as the model received it, and last_response shows the value the request used.
Secrets page shows who last changed each value
The Secrets page now shows who last changed each secret’s value in the environment you are in, alongside when.
- Open Settings > Secrets and read the Last editor column to see the member who made the change.
- A change made through a personal key names the member who created the key.
- Service account appears when a service account, such as an Atlas routine, made the change.
- The column stays empty when no editor is known.
When values move between environments by publishing, merging, cloning, discarding, reverting or pulling from Main, each secret keeps its original editor and time. A change made through the API appears in open builder tabs straight away.