> For a complete page index, fetch https://docs.synthflow.ai/llms.txt. For full documentation content, fetch https://docs.synthflow.ai/llms-full.txt.

# Changelog

## Manage number pools in the app

> Manage an agent's phone numbers in the app, and load an earlier version into your draft without going live.

![Phone tab of Deployment Settings listing three numbers, the first marked Default, above the Add Phone Number button](https://storage.googleapis.com/granular-changelog/doc-images/changelog_number_pooling.png)

One agent can now use several phone numbers, and you manage them all from its deployment settings. Add or remove numbers there, and pick which number calls you when you test a call. See [Number Pooling](/number-pooling).

## Revert a version into your draft

![Revert Version menu offering Revert, which loads the version into the draft, and Revert and publish, which makes it live as a new version](https://storage.googleapis.com/granular-changelog/doc-images/changelog_revert.png)

If a revert put the wrong configuration live, you had to publish again to undo it. Restoring a version now offers **Revert**, which loads it into your draft so you can test and edit it first. **Revert and publish** still makes it live right away. See [Restore this version](/version-control#restore-this-version).

## See unpublished changes anytime

> The Changes panel lists everything your draft changes before you publish it.

![Changes panel next to the agent editor, opened from the Draft · 1 change tag, showing a one-line prompt edit](https://storage.googleapis.com/granular-changelog/doc-images/changelog_changes_drawer.png)

If you lose track of what you edited since the last publish, you no longer have to open **Publish** or **Compare changes** to find out. Select **Draft · N changes** in the agent editor header to open the [Changes panel](/version-control#draft-changes). Select a setting's name in the panel to jump to it in **Global Settings**.

## New German voices

> Ten native German ElevenLabs voices join the recommended voices, under a new German accent filter.

If you build agents that speak German, you now have ten more native German voices to choose from. They are in the recommended voices, with a new **German** accent that marks native German speakers.

In the [voice list](/voice-configuration#voice), set the language to **German** and the accent to **German** to find Ben, Heidi, Susi, and the seven others.

## Compare any two agent versions

> Compare any two versions of an agent, including the draft, to see exactly what changed between them.

![Compare changes dialog comparing V2 with the live V4, showing a one-line prompt edit with the removed text in red and the added text in green](https://storage.googleapis.com/granular-changelog/doc-images/changelog_versioning_compare_changes.png)

Until now, you could only see what changed between a version and the one right before it. Now you can compare any two versions of an agent, including your draft.

Select **Compare changes** in the agent's [version history](/version-control#compare-changes) to pick the two versions.

## Listen to transfer recordings

> Play recordings of warm transfers, turn on expressive delivery, and use new REST API and MCP server tools.

![Actions tab of Call Details with a warm transfer action marked Success, showing a Transfer recording player with playback speed and download buttons](https://storage.googleapis.com/granular-changelog/doc-images/changelog_recording_transfers.png)

When a warm transfer failed, you only had the caller's recording. With [Save Recordings](/security-and-compliance#save-recordings) and [Continue recording after the transfer](/call-transfers#continue-recording-after-transfer) on, warm transfers now get their own recording, including ones that failed after they were answered.

## List agent versions via API

An agent's [version history](/version-control#manage-previous-versions) could only be viewed in the app. Through the API, [List agent versions](/api-reference/platform-api/agents/list-agent-versions) now returns every published version, newest first, with who published it, when, and which version is live.

## Expressive delivery with one toggle

Getting an agent to sound warm or reassuring meant writing audio tags into the prompt. Turn on [Expressive mode](/voice-configuration#expressive-mode) and the agent picks its own delivery, and transcripts stay free of tags.

## New MCP server tools

The [MCP server](/mcp-server#supported-tools) can now read a single knowledge base item and create post-call email actions. `get_agent` also returns the agent's business hours.

## Call recordings move to a new endpoint

> Switch to the new recordings endpoint before recording_url becomes unavailable on January 10, 2027, and reconnect Salesforce by November 30, 2026.

The `recording_url` field in the [calls endpoints](/api-reference/platform-api/calls/get-phone-call) becomes unavailable on **January 10, 2027**. It keeps working until then. Use [Get a recording](/api-reference/platform-api/recordings/get-call-recording) instead: ask for a download link whenever you need the audio, instead of storing one.

Follow our [guide](/recordings-migration-guide) to switch.

## Reconnect Salesforce by November 30

We changed how Synthflow connects to Salesforce. To keep your Salesforce custom actions working, reconnect Salesforce in each workspace by **November 30, 2026**.

[Follow these instructions](/integrate-with-salesforce#use-salesforce-in-custom-actions) to reconnect.

## Webhook signatures for subaccounts

> Share your workspace's webhook secret key so subaccounts sign their webhooks too.

![Webhook signature settings with the option to share the secret key with subaccounts](https://storage.googleapis.com/granular-changelog/doc-images/changelog_webhook_signature_subaccounts.png)

Subaccount webhooks went out unsigned, so you couldn't confirm they came from Synthflow. You can now [share your workspace's secret key with your subaccounts](/security#subaccounts), so their webhooks carry the same `HTTP_SYNTHFLOW_SIGNATURE` header you already validate.

## Manage subaccounts from Settings

> The sidebar no longer has an Agency button. Subaccounts and other agency tools now live under Settings.

![Redesigned main sidebar without the Agency button, with Settings as the place to manage subaccounts](https://storage.googleapis.com/granular-changelog/doc-images/changelog_agency_button.png)

We redesigned the main sidebar to cut clutter and keep it focused on the pages you use every day. As part of that change, the **Agency** button is gone and everything it opened now lives in **Settings**. To manage your [subaccounts](/subaccounts), select **Settings**, then **Subaccounts**.

## Filter call logs by action

> Find every call that ran a specific action, and set action inputs and advanced transfer settings via API.

![Action filter on Call Logs open on All Actions, with a submenu for each action type such as Real-Time Booking, Call Transfers, and Information Extractors](https://storage.googleapis.com/granular-changelog/doc-images/changelog_filter_logs_by_action.png)

To check how an action performed, you had to open calls one by one. In [Call logs](/logs#action-filter), the **Action** filter now shows only the calls that ran the action you pick.

Starting from the action works too: on the **Actions** page, open an action's menu and select **View Call Logs** to see its calls.

Through the API, [List calls](/api-reference/platform-api/calls/list-calls) filters by action too.

## Set action input sources via API

Actions attached to an agent through the API asked the caller for every input, even values you already had. When you [attach an action](/api-reference/platform-api/actions/attach-action), you can now choose where each input comes from, with the same [input sources](/variables#choose-how-an-input-variable-gets-its-value) as in the app.

## Advanced transfer settings via API

The [advanced call transfer settings](/call-transfers#advanced-settings) could only be set in the app. You can now set them when you [create](/api-reference/platform-api/actions/create-action) or [update](/api-reference/platform-api/actions/update-action) a call transfer action.

## See every variable on a call

> A new Variables tab in call logs shows the value of every variable a call used, and the API now lists and updates phone book entries.

![Variables tab in Call Details with the filter, Collapse all and Copy as JSON controls, and the System group listing call\_id and user\_phone\_number](https://storage.googleapis.com/granular-changelog/doc-images/changelog_variables_drawer.png)

If you've been piecing together what a call sent to your actions, open the new **Variables** tab in [Call logs](/logs#call-logs). It shows the value each variable had on that call:

* [System, pre-call, and collected variables](/variables), and inbound SIP headers
* Every action's inputs and outputs, including transfer SIP headers

## Manage phone book entries via API

Once you added numbers to a phone book through the API, you could not read them back. You can now [list a phone book's entries](/api-reference/platform-api/phone-books/list-phone-book-entries), up to 100 per page, and [update an entry](/api-reference/platform-api/phone-books/update-a-phone-book-entry) by sending its new `phone_number` and `transfer_condition`.

## Redesigned sidebar

> A sidebar organized around BELL with the changelog built in, plus full knowledge base sources and post-call email actions in the API.

![Synthflow sidebar with Agents, Knowledge Base, Actions and Memory under Build, Test Center under Evaluate, and Phone Numbers, Campaigns and Workflows under Launch](https://storage.googleapis.com/granular-changelog/doc-images/changelog_redesigned_sidebar.png)

Every page in the sidebar now sits under the step of the [BELL framework](/getting-started#the-bell-framework) it belongs to.

**Help** and **Changelog** have moved to the bottom of the sidebar. Open **Changelog** to see what's new without leaving the app.

## Read a full knowledge base source via API

Listing a knowledge base's sources returns only the first 200 characters of each text source, and no content for other types. To read one source in full, use [Get a knowledge base source](/api-reference/platform-api/knowledge-bases/get-knowledge-base-source). For a text source, `content` holds the full text. For a file, `url` is a download link that expires after 7 days.

## Send emails via API

[Post-call email actions](/post-call-email) showed up in the API, but you could only create them in the app. You can now [create](/api-reference/platform-api/actions/create-action) and [update](/api-reference/platform-api/actions/update-action) them with `SEND_EMAIL`, sending a `name` and 1 to 50 `recipients`.

## GPT-6 Luna

> Run your agents on OpenAI's GPT-6 Luna, set business hours through the API, and pick custom action results without Initialize.

![AI Model dropdown with GPT-6-Luna selected, below GPT-5.4, GPT-5.4-Mini and GPT-5.4-Nano](https://storage.googleapis.com/granular-changelog/doc-images/changelog_gpt_6_luna.png)

You can now run your agents on OpenAI's GPT-6. Pick **GPT-6-Luna** in the [AI Model selector](/general-configuration), or set `llm` to `gpt-6-luna` when you [create](/api-reference/platform-api/agents/create-assistant) or [update](/api-reference/platform-api/agents/update-assistant) an agent.

## Business hours through the API

An agent's [business hours](/business-hours) could only be set in the app. You can now set them through the API as well. When you [create](/api-reference/platform-api/agents/create-assistant) or [update](/api-reference/platform-api/agents/update-assistant) an agent, send:

* `working_hours`: the weekly schedule.
* `out_hours_action`: the [call transfer](/call-transfers) or custom action that handles inbound calls outside those hours.

Getting or listing agents returns both.

## Custom action results through the API

Until now, you had to initialize a [custom action](/about-custom-actions) before you could pick its result variables. Through the API, you can now enter the response yourself instead. When you [create](/api-reference/platform-api/actions/create-action) or [update](/api-reference/platform-api/actions/update-action) the action, send:

* `api_response`: an example of the response your API returns.
* `variables`: the fields to keep from it, such as `results.data.order_id`.

## Redesigned call Overview tab

> See why a call ended, whether it reached voicemail, and how each transfer went, right on the call.

![Call overview with end reason, disconnect reason, voicemail result, and the Transfer section](https://storage.googleapis.com/granular-changelog/doc-images/changelog_call_overview_detail.png)

The post-call webhook was one place to see why a call ended or whether it reached voicemail. The **Overview** tab in [Call logs](/logs#call-logs) now shows it on the call itself:

* The summary, end reason, disconnect reason, and voicemail result, side by side. Hover over a field to see what it means.
* **Transfer**: each attempt, its outcome, and how far into the call it started, so you no longer search the transcript to see if a transfer worked.
* **Call**, **Agent**, **Contact**, and **Technical** groups for the rest, including the agent number, the agents used, and the node timings.
* **Copy** menu in the header: copy the call ID, a link to the call, or the full call data as JSON for a support ticket.

## ElevenLabs v4 and v4 Turbo

> Pick ElevenLabs v4 or v4 Turbo as your voice model, get every action run in the Calls API, and filter voices by accent.

![Voice Model dropdown with ElevenLabs v4 selected, above ElevenLabs v4 Turbo and Deepgram Aura-2](https://storage.googleapis.com/granular-changelog/doc-images/changelog_elevenlabs_v4.png)

You can now pick **ElevenLabs v4** or **ElevenLabs v4 Turbo** as your [voice model](/voice-configuration#elevenlabs-v4-and-v4-turbo). Both support [audio tags](/voice-configuration#audio-tags), the same as V3.

In the API, set `voice_synthesizer` to `eleven_v4` or `eleven_v4_turbo`.

## Every action run in the Calls API

When a call ran the same action more than once (such as knowledge base lookups or custom actions), the Calls API returned only the last run. Add `include_all_actions=true` to [Get a call](/api-reference/platform-api/calls/get-phone-call) or [List calls](/api-reference/platform-api/calls/list-calls) to get `all_executed_actions`, which lists every run with its start and finish time. `executed_actions` is unchanged, so existing integrations keep working.

## Voice accent in GET /voices

You can now choose a voice by accent without opening the app. [GET /voices](/api-reference/platform-api/voices/get-voices) returns each voice's `accent`, such as `american`.

## Transfer results in Call logs

> See transfer results in Call logs, and follow each warm transfer attempt in the SIP ladder.

![Call logs table with the Transfers column showing one attempt per call, marked Success or Failed](https://storage.googleapis.com/granular-changelog/doc-images/changelog_transfers_column.png)

Check whether [transfers](/call-transfers) connected without opening the call. Select the arrow in the **Transfers** column to see every attempt and its result.

## SIP ladder for warm transfers

![SIP Ladder with Warm transfer 1 selected, showing the INVITE from Synthflow to Twilio and the 100 response](https://storage.googleapis.com/granular-changelog/doc-images/changelog_sip_ladder_transfers.png)

The [SIP ladder](/sip-call-ladder#warm-transfers) now shows every warm transfer attempt, not only the original call, so you can follow each transfer's SIP signaling message by message, with timestamps.

## The new Synthflow app is coming to everyone

> Every workspace moves to the new Synthflow app on October 14, 2026, with no action required.

![The new Synthflow app at app.synthflow.ai](https://storage.googleapis.com/granular-changelog/doc-gifs/appsynthflowai_compressed.gif)

On **October 14, 2026**, every workspace moves to the faster [new Synthflow app](/new-synthflow-app) at `app.synthflow.ai`. Your data and integrations will be right where you left them, so **no action is required**.

This does not apply to [custom domains](/custom-domain). Check out the [guide](/new-synthflow-app) to see what changes and what remains.

## Actions API: Readable names

> Action names keep the text you type, and Get an agent returns more fields.

Names of call transfer, post-call SMS, in-call SMS, and WhatsApp actions are now saved exactly as you type them, up to 64 characters, through both the [Actions API](/api-reference/platform-api/actions/create-action) and the app.

## Agents API: Updated Get an agent response

The [Get an agent](/api-reference/platform-api/agents/get-assistant) response now includes these fields:

* `knowledge_base_id`
* `agent.greeting_message_mode`
* `agent.greeting_message_human_talk_timeout`
* `agent.suppression_level`
* `agent.redact_pii`
* `agent.voice_style`, now returned as a decimal instead of `0`

## Improved performance in Logs search

> Call logs and webhook logs search faster, with exact matches on phone numbers and IDs.

Searching your logs is now much faster, including in high-volume workspaces.

* **Call logs:** The **Phone Number** filter now only finds exact matches, so results load instantly. Enter the full number, including the country code.
* **Webhook logs:** Search by a full Call ID, webhook log ID, or action ID for instant results.

## Extended list of transfer failure reasons

> Failed transfers now report nine specific failure reasons instead of a generic one.

Failed transfers now report more specific failure reasons:

* `number_not_found`
* `number_incomplete`
* `destination_unavailable`
* `rejected_by_destination`
* `transfer_blocked_by_carrier`
* `carrier_unavailable`
* `routing_loop`
* `media_negotiation_failed`
* `transfer_method_unsupported`

Find every state, reason, and SIP code in the [transfer outcome states reference](/transfer-outcome-states).

## Prompt Reuse metric

> Prompt Reuse in Analytics shows how much of its prompt an agent reuses, which lowers latency.

![Agents Performance card on the prompt reuse tab, showing one agent at 74% and another at 3%](https://storage.googleapis.com/granular-changelog/doc-images/prompt_cache.png)

**Prompt Reuse** is a new metric in [Analytics](/collect-analytics) that shows the health of your agent's prompt. It measures how much of the prompt the AI model reused instead of reading it again.

Use it to understand why some agents respond faster than others: the more of its prompt an agent reuses, the lower its latency in the call.

Learn more about [Prompt Reuse](/collect-analytics#prompt-reuse).

_Showing the 20 most recent of 112 entries. Append `/llms.txt` to the changelog URL for the complete index._