Skip to main content

October 12, 2026

View-only teammates

Give a teammate the viewer role on the Users page and they can see everything on the dashboard and change nothing: calls, transcripts, analytics, agents and tests, all readable, with every change refused (403 viewer_role). Made for people who review and report. Roles are set by admins; ask us if the viewer role does not appear in your role list yet.

Guardrails

Keep an agent on its job whatever the person says: Stay on purpose (declines unrelated requests and returns to its purpose), Resist manipulation (keeps its persona and instructions when someone claims authority, says it is a test, or asks it to play another role), and Never discuss (subjects it declines outright). Choose what happens when the person keeps pushing: keep steering back, or say goodbye and end the call. They apply on calls, in tests, over SMS and in web chat, and are saved with the agent’s versions. On the agent’s Behavior section, or guardrails on the agent. Guide

Website sources up to 500 pages

A website knowledge source now reads up to 500 pages (was 50), starts from the site’s sitemap so pages no link leads to are included, and reads several pages at once. Guide

Tests with canned tool responses

A test suite or a single test can pin what the agent’s custom tools return (toolMocks), so a booking or lookup scenario runs against a known answer and never reaches your live systems: the webhook is not called and the agent gets the pinned response, with the HTTP status you choose. A test’s entries win over the suite’s; tools without an entry are called for real. Works in chat, SMS and web chat tests (voice tests are next), and mocked calls are marked in the transcript. On the dashboard: Tool responses on the suite and on each test. Guide

October 11, 2026

Right-to-left languages read correctly

Transcripts, summaries, conversation threads and the website chat widget now follow the direction of the text, so Arabic and Hebrew read right to left.

Read-only API keys

When creating an API key, choose Full access or Read-only. A read-only key can read calls, agents, analytics and everything else, and is refused on any request that would create, change or delete (403 read_only_key). Made for reporting, dashboards and integrations that only read. Reference

Calls record their language

Each call now records the language it ran in: filter the Calls list by language, see it on the call page, and read it as language on the call record (GET /calls?language=es).

Activity across the account

A new Activity page (under Monitor) lists every change to your agents, newest first: who made it (a teammate or an API key), when, which agent, and which settings changed, with filters for range, agent and person. Each row opens the agent’s History, where the full before and after can be read and a version restored. Home shows the latest five. Also available as GET /activity. Reference

Agent activity at a glance

The Agents list shows each agent’s calls over the last 7 days (connected, failed and typical length on hover) and when its last call was. The Home page lists the busiest agents for the selected range, each opening that agent’s Insights. Each agent’s page has an Insights section with calls, cost, duration, transfers, reply time and voicemail rate against the period before, a trend that opens the calls behind any point, and call outcomes with a one-click way to turn failing calls into tests. Also available as GET /analytics/agents. Reference

Custom tools without hand-written JSON

A tool’s parameters are now edited as a form: name, type, description, whether it is required, and allowed values. The JSON view stays for nested schemas. Deleting a tool asks first.

October 10, 2026

More female Studio voices

Three new American Studio voices: Audrey (bright, warm and professional), Natalie (calm, reassuring and confident) and Chloe (friendly, bright and warm). The voice list now shows each voice’s gender, accent and tone, filters by gender, and lists female voices first. GET /voices returns accent and description for Studio voices.

Test calls straight from the browser

The test panel’s Talk button now connects your browser to the agent directly: clearer audio and a faster line than a phone connection. Transfers are simulated during a test, so nobody is dialed. Also available as POST /agents/{id}/test-call.

Describe an agent and get a ready prompt

On Create agent, choose Describe it and say what the agent should do. You get a name, an opening line and a prompt organized for phone calls to review, with transfers from any number you mention and a list of facts to add. Also available as POST /agents/draft. Guide

Live calls in the dashboard

A new Live page shows every call in progress with its transcript as it happens. Listen in, guide the agent, or end a call from the same place. Guide

Live instructions survive a talk-over

A Say this now instruction that the caller talks over is brought up again in the agent’s next reply, until it has been said in full. The dashboard’s live controls now read Say this now and Keep in mind, each with a note on what it does.

The agent keeps listening after its goodbye

If the caller says something new while the agent is wrapping up, the agent answers it and the call continues. A goodbye from the caller still ends the call, and once the caller and the agent have both said goodbye, the call ends right away.

A “hello” during the opening line gets no second reply

When the person who answers says “Hello?” while the agent’s opening line is playing, the agent finishes its opening and waits for their answer instead of replying to the hello as well. A question or a “yes” said over the opening is still answered.

Transcripts keep the last words of a call

When the caller hangs up while the agent is speaking, the transcript keeps what the agent said up to that moment, marked as cut off, and the caller’s own last words.

Callers heard from their first word

On inbound calls, a caller who starts talking during the agent’s greeting is heard from the first word, so the agent does not ask again for details the caller already gave.

Short answers in other languages

Agents set to Spanish, French, German, Italian, Portuguese, Dutch, Russian, Hindi or Japanese recognize one-word answers such as “sí”, “oui” or “ja” more reliably. Multilingual agents change voice and language only on a full sentence. Guide

Check-ins when the line goes quiet

Agents on RevRing Max always say their check-in when the caller goes quiet, and keep listening for the caller’s reply.

Acknowledgments never stop the agent

“Yes”, “sure”, “exactly”, “sounds good” and similar words said while the agent is speaking no longer pause or restart it; “wait”, “stop” or a question still interrupt at once. Guide

Warm transfers to a line that answers in silence

When the person you transfer to picks up without speaking, the agent introduces the caller right away, and a line where nobody ever speaks is given up after a minute with the caller told, instead of holding them.

Call screening with a pause before speaking

When a call screening assistant answers during the agent’s pause before its opening, the agent answers the screener once and does not also play its opening.

Reply time on every turn

Agent turns in the transcript carry latency.replyMs, what the caller waited for that reply, and metrics.latency.stop_to_audio summarizes it per call. Guide

Backchannels (beta)

Agents can add a short “mm-hmm” during long caller turns. Off by default. Guide

Inbound lines held for callers

An inbound reserve keeps lines that outbound calls (API calls, campaigns, tests) never take, so callers and website visitors get through while a dialer runs. Guide

Clearer reasons when a call fails

errorMessage is a short code or a plain sentence everywhere: the API, the dashboard and webhooks. Guide

Chat with an agent

Test an agent by typing, in the dashboard’s test panel or with POST /agents/{id}/test-chat. Guide

Reply times

GET /analytics/reply-times reports how fast an agent answers over its recent calls, and the agent page shows it.

Studio voices have new names

The built-in Studio voices are now Maya, Evan, Lena, Caleb, Grant, Wesley, Bennett, Rowan, Imogen, Harriet, Oliver, Hugh and Lachlan. Voice IDs are unchanged and every agent keeps its voice.

A new dashboard

A cleaner look with a dark theme, faster pages, search with ⌘K, an agent builder organized by section with one save button and a warning before you leave unsaved changes, templates for new agents, and a call page that explains why a call failed.

October 8, 2026

The agent speaks while a tool runs

Give a custom tool an executionMessage such as “Hang tight, I’m pulling up your order” and the agent says it while a slow webhook works, then answers from the real result. Fast responses get no line. In the dashboard, open the tool and fill in Say While It Runs. Agents imported from Retell AI and ElevenLabs keep their execution speech. Guide

Call variables in tool requests, and saving response fields

Use {{name}} in a tool’s URL path, query values and header values, and save fields of a tool’s JSON response as call variables with responseVariables for the tools that run next. In the dashboard: Save Response Fields as Variables. Guide

Silent tools and retries

speakAfterResult: false makes a tool run without the agent commenting on its result. maxRetries (0 to 3) retries a lookup after a connection error or a 429, 502, 503 or 504 response, within the tool’s time limit. Guide

Calls that never connect fail fast

An outbound call that cannot be picked up by an agent within seconds is retried automatically, and if it still cannot be placed it fails right away with agent_unavailable instead of staying in progress for up to 30 minutes.

Voicemail

A recognised voicemail keeps the outcome VOICEMAIL_DETECTED even when the mailbox hangs up before a message can be left, hang-ups after a message are faster, and call screening assistants are answered with one short line and held quietly until the person or the voicemail prompt.

October 7, 2026

Talk to the agent during a call

Send the agent an instruction while a call is live with POST /v1/calls/{id}/instructions: have it say something now ("speak": true) or shape its next reply. Instructions appear in the transcript, and the MCP server has a matching tool. Guide

Timeout per tool

Set timeoutSeconds (1 to 120) on a custom tool to give a slow webhook more time; the default stays 8 seconds. Guide

Hand off to another agent

An agent can pass the caller to another agent on the same call, with its own prompt, tools and voice, using handoffRoutes. The caller never redials, and the other agent picks up with the whole conversation. Chat tests follow handoffs too. Guide

Turn call recording off

Set recordingEnabled: false on an agent (or turn off Record calls on its Advanced tab) and no audio is recorded for its calls; transcripts and analysis are kept as usual. Guide

Failed tool calls are marked

Tool calls in call transcripts carry isError: true when the tool failed or timed out, so your systems and alerts can tell them apart from normal results.

Pronunciation

Tell an agent how to say names, brands and abbreviations with pronunciations (for example Nguyen said as “Win”). Only the voice changes; transcripts keep the word as written. Agents imported from ElevenLabs bring their pronunciation dictionaries’ alias rules. Guide

Caller keypad input

Turn on keypadInput and digits the caller presses, such as a PIN or an account number, reach the agent as their answer, with an optional finish key, digit limit and pause. Agents imported from Retell AI keep it on. Guide

The voice follows a multilingual caller

A multilingual agent with language presets now switches to a language’s preset voice when the caller changes language mid-call, and back to its own voice for other languages. Guide

MCP servers on agents

Give an agent the tools of up to 5 MCP servers, such as your CRM, helpdesk or internal systems, next to its custom tools. Add them on the agent’s Tools tab (with Test connection to pick the tools it may use) or with mcpServers on the agent. Tools load when each call, chat or text starts and never delay the call. Agents imported from Retell AI or ElevenLabs bring their MCP servers. Guide

Per-call content, your own data, and safe retries

POST /v1/calls now takes firstMessage, promptAppend, maxDurationSeconds and voicemailMessage for a single call, metadata (your own data, up to 16 KB, stored on the call and sent in its webhooks), an Idempotency-Key so a retried request never dials twice, and honorDoNotCall to refuse numbers on your do-not-call list. Guide

Voicemail rate and call filters

The dashboard Overview shows your voicemail rate (answered outbound calls that reached a voicemail box), and GET /v1/calls and the CSV export filter by hang-up cause, for example ?hangupCause=VOICEMAIL_DETECTED. Guide

October 6, 2026

Live booking on your calendar

Connect Cal.com or GoHighLevel (beta) on the new Integrations page (or POST /v1/integrations), then turn on Appointment booking on an agent’s Tools tab (or set booking). The agent gets check_availability and book_appointment: it offers only the times your calendar returns, confirms the one the caller picks, and books it live on calls and in chats. Calendar keys stay on our side and never reach the agent. Guide

Contact memory

Agents can now remember the people they talk to. With contactMemory (or Contact memory on the agent’s Numbers tab), each call adds what matters for next time to a short memory kept for the person’s phone number: the name they go by, what they want, appointments with dates, concerns and next steps, never card numbers or IDs. The next call, inbound or outbound, starts with it. See, correct or erase memories with /v1/contacts. Off by default. Guide

Callbacks, recognized

When someone you called in the last 30 days calls your agent’s number back, the agent knows it is a callback: the outbound call’s variables (their name, your lead ID) come with the call, the agent is told how that call went and what was said, and it greets a returning caller instead of repeating an opening written for calls it places. Callbacks carry is_callback, callback_call_id, last_call_time and last_call_outcome, and your pre-call webhook gets a callback object. On by default; tune it with inboundCallbacks (window and an optional callback greeting) or on the agent’s Numbers tab. Guide

Do-not-call requests, handled automatically

Turn on doNotCallDetection (or Detect do-not-call requests on an agent’s Advanced tab): when a caller asks not to be called again, the call shows doNotCallRequested: true, the post-call webhook carries it, and their number joins your do-not-call list so campaigns never dial it. Declining an offer or asking for a call back later does not count. Guide

A/B test agent versions

Split an inbound number’s calls between versions of an agent with trafficSplit (2 to 5 environments with weights), then compare them with GET /v1/analytics/versions: failure rate, call length, negative sentiment and success-criteria pass rates per version. Guide

Response latency on every call

Each call’s metrics.latency now shows how quickly the agent answered: the caller’s wait from the end of their speech to the agent’s reply, broken into turn detection, transcription, thinking and voice, each as p50, p90 and max across the call’s turns. Guide

One agent, many languages

Pass language when you place a call (or from your pre-call webhook for inbound calls) to run that call in the caller’s language, and give the agent languagePresets: a voice and opening line per language, used whenever a call runs in that language. Agents imported from ElevenLabs bring their per-language opening lines. Guide

Alerts

Hear about problems before your customers do. Create alerts on failed calls, calls failing a success criterion, negative caller sentiment, tool errors, undelivered post-call webhooks or call volume, for one agent or all, over 15 minutes to 24 hours. Each alert sends one webhook or Slack message when it fires and one when it resolves, with a test button and a history of every notification. Open Alerts in the dashboard or use /v1/alert-rules. Guide

Knowledge from whole websites, kept current

A URL source can now read a whole help center or documentation site: add crawl with maxPages (up to 50) and an optional includePath, and RevRing follows the site’s links under that path into one source. Set refreshDays to read the pages again on a schedule, or refresh a source on demand with POST /v1/knowledge-bases/{id}/sources/{sourceId}/refresh. Agents keep answering from the current content during a refresh, and new content replaces it only once it is fully indexed. Guide

Caller sentiment, re-run analysis and call export

Every call with caller speech now carries sentiment (positive, neutral, negative or unknown) on the call and in the post-call webhook; Retell-format webhooks fill user_sentiment from it. After changing an agent’s extraction fields or success criteria, re-run the analysis on past calls (POST /v1/calls/{id}/reanalyze, up to 50 at once with POST /v1/calls/reanalyze, or Re-run analysis on a call). Download call history as CSV with sentiment, variables, extracted data and evaluation results (GET /v1/calls/export, or Export CSV on Calls). Guide

Transfer events on your webhook

Add transfer.started, transfer.bridged and transfer.cancelled to an agent’s webhookEvents (or turn on Also send transfer events) to hear about transfers as they happen: started (warm or cold, with the destination), connected, or not completed (declined, voicemail or failed). Agents using the Retell compatibility format receive Retell’s transfer_started, transfer_bridged and transfer_cancelled. Guide

Listen to calls live

Open any call in progress in the dashboard and select Listen to hear both sides as it happens. Nobody on the call can hear you or knows you are listening, so the call is never affected. The transcript on the same page updates every few seconds while the call runs.

Remove personal details from call transcripts

Set an agent’s storageMode to redact_pii (or choose it under Advanced > Call data storage) and names, phone numbers, emails, addresses, dates of birth, ID, card, account and medical record numbers, and passwords are replaced with labels like [phone_number] in every call’s transcript and summary, before the call is stored or sent to your webhook. Pick categories with piiCategories. Calls show piiRedacted: true. Agents imported from Retell AI or ElevenLabs carry their PII settings over. Guide

October 5, 2026

Connect Twilio in one step

Bring Twilio numbers over with your Account SID and Auth Token: RevRing creates the Elastic SIP trunk, points it here, moves the numbers you pick and can assign them to an agent. From the dashboard (SIP Trunks > Connect Twilio) or POST /v1/sip-trunks/connect/twilio. Guide

Fix failing calls

Build a test suite from an agent’s recent calls that failed a success criterion and run it, in one step from the dashboard Overview or POST /v1/test-suites/from-failing-calls. Success criteria now count a skipped required action as fail; an outcome the call never reached stays unknown. Guide

Create a test from a real call

One click on a call (or POST /v1/test-suites/{id}/tests/from-call) turns it into a test case: the caller’s side becomes the script, the failed success criteria the rubric. Guide

Call started event

Set webhookEvents: ["call.started"] on an agent to get a webhook when each call connects, in RevRing’s shape or Retell’s call_started. Guide

Route inbound calls from your pre-call webhook

The pre-call webhook’s answer can now pick the agent or environment for the call, override the opening line, voice or prompt, or reject the call. Retell and ElevenLabs inbound webhook answers work unchanged. Guide

Manage RevRing from Claude, Cursor and VS Code

A management MCP server at https://api.revring.ai/v1/mcp lets AI tools list and edit agents, place and end calls, and read transcripts and results with your API key. Guide

Keep your webhook handler when you switch

Set postCallWebhookFormat to retell or elevenlabs and the post-call webhook arrives in that platform’s payload shape, filled from the RevRing call record. The importer sets it automatically. Guide

Knowledge bases hold 50 sources

The limit per knowledge base is now 50 sources (was 10).

Campaigns

Call a contact list with one agent. Contacts without a time zone are called in the local time of their number’s area code, and voicemailStrategy: "last_attempt" leaves a voicemail only on the final attempt. Calling windows follow each contact’s own time zone, retries happen automatically by outcome (no answer, busy, voicemail, declined, failed), new calls are paced to suit your carrier, numbers on your do-not-call list are never dialed, and a CSV export carries every contact’s outcome with extraction and evaluation columns. Campaign calls carry campaignId on the call record and in the post-call webhook. Guide

Import agents from Retell AI and ElevenLabs

Bring existing voice agents over with your API key for the other platform. A dry run reports how every setting carries over (prompt, opening, variables, tools, transfers, voicemail, knowledge, post-call analysis) and what is left to you; importing creates the agents with their tools and knowledge bases, and importing again updates them. Guide

End a live call from your server

POST /v1/calls/{id}/end hangs up a call in progress immediately; it is recorded as API_ENDED and still gets its recording, summary and post-call webhook. Use it when your own logic decides the call should stop. Reference

Concurrency headroom

GET /v1/analytics/active now also returns concurrency (active, limit, available), counted exactly like the check behind concurrency_limit, so dialers can pace calls instead of retrying. Reference

Spoken text never reads a placeholder aloud

If an opening line, voicemail message, hold or intro message references a {{variable}} that has no value on a call, the placeholder is dropped and the sentence tidied (“Hi , this is Maya” becomes “Hi, this is Maya”) instead of being read out.

Webhook signing

Every webhook RevRing sends (pre-call, post-call, message received) can now be signed. Create a secret with POST /v1/webhooks/signing-secret and verify the X-RevRing-Signature header on the raw body. Opt-in per organization; nothing changes until you create a secret. Guide

Webhook retries

A post-call webhook that fails (connection error, timeout, non-2xx) is retried 1, 5 and 30 minutes later. Handlers should be idempotent on the call id. The outcome is on the call record as postCallWebhookStatus and postCallWebhookAttempts. Guide

Test runs against a live version

Test runs use the saved draft by default. Pass environment (production or an environment name) on POST /test-suites/{id}/runs to test the version callers get. Guide

October 4, 2026

Drafts, publishing and environments

Agents can now be edited without going live. Switch an agent to manual publishing (publishMode: "manual"), edit the draft, test it, and publish when ready; earlier versions can be made live in one call. Named environments pin a version with their own variable defaults and can be bound to inbound numbers, the web widget and POST /calls. Every call record carries the version it ran on. Guide

Post-call data extraction

Define typed fields per agent (postCallExtraction) and get them filled from each conversation on the call record and in the post-call webhook as extraction. Guide

Post-call evaluation

Write success criteria per agent (postCallEvaluation) and every call is judged pass, fail or unknown with a one-sentence reason. Pass rates per criterion are on the analytics overview and the dashboard. Guide

SIP trunk address validation

A trunk’s proxy must now be a hostname or IP that resolves. Typos such as a misspelled carrier domain are refused when the trunk is created instead of producing a trunk that can never place a call. Reference

Noise suppression for every agent

Background noise and other voices on the caller’s side are suppressed before the agent listens. On by default for all agents; turn it off per agent with noiseCancellation: "off".