Dapta Changelog

A Warm Transfer can play a message while the caller waits for a person. Turn on Message while the caller waits, write up to 300 characters and choose how often it plays, from every 30 seconds to every 5 minutes. The hold music pauses, the agent says the message and the music resumes. It's off by default and applies to phone calls.

A voice agent can have more than one Dapta Calendars Scheduling action, each with its own meeting, time zone and instructions, so one agent can book, say, a demo and a follow-up visit. The agent knows which meeting each action is for. Before, adding a second action replaced the first.

In conversations on WhatsApp numbers connected through Meta, the + menu of the Inbox reply box offers Image, File and Audio, so you can send a picture, a document or a recorded voice note. Images go up to 5 MB and other files up to 16 MB, in common formats such as JPEG, PNG, PDF, Word, Excel, MP3 and MP4. Files the AI agent sends now show as their own cards.

The notifications bell opens on a What's new tab with Dapta's latest releases, each marked as a new feature, improvement, fix or beta, with its date. The All tab groups your notifications by date and has an Unread switch to show only the ones you haven't opened. Both tabs count what you haven't read, and See latest news opens the news page in the docs.

The CRM has a New Contact button: enter a name and a phone number with its country code, and save. In a contact's panel, Edit turns the name, phone and every custom column into fields you change in place, and action cards let you call, send a template, add the contact to a voice or text sequence, or write a note. These need permission to edit the CRM.

On a call's detail page, the badge under the player shows whether the call met its goal. Owners, admins and members can switch it between Goal met and Goal not met, with a reason of at least 10 characters. The Call result window keeps the AI's original verdict and the history of changes, and call logs, their filter and the agent's success metrics use the corrected one.

A text sequence's Conversations tab has Add contact, to pick a contact from the Dapta CRM, and Import file, for a CSV or XLSX with the columns the sequence uses. Numbers already in the sequence are skipped, and the import reports how many were added, already there or invalid. Contacts can't be added once a sequence has ended, so use one with no end date to keep loading them.

The Dapta MCP server has a list_voices tool that lists the voices the app's voice picker offers your workspace, including your own cloned voices, each with a sample to listen to. Search by name, language, gender or accent. Creating or updating a voice agent with a voice outside that list is refused, and the closest matches are suggested.

Text and voice agents can run on Dapta Auto, a model Dapta chooses, tests and keeps up to date. New text agents start on it, at 40 credits per message, and it reads images and voice notes. On a voice agent, pick it under Voice Settings, LLM Model, and calls keep the same per-minute rate. Existing agents keep their model until you switch them.

Carousel templates can be sent from the Inbox, voice and text sequences, a voice agent's Send WhatsApp template function and Flow Studio. Each card takes its own variables and, if you want, its own image or video for that send, from a public link on the same server as the template's sample file. Without one, the card uses the sample saved with the template. In the editor, each card's file can be uploaded or given as a public URL.

A Use Dapta from card in the sidebar opens a new MCP page that starts with signing in with your Dapta account, no API key needed. Pick ChatGPT, Codex, Claude, Claude Code or Cursor and follow the steps and short video for that app, and Cursor has an Add to Cursor button. Any workspace member can connect, and API keys move to an Advanced section.

Custom workspace roles set Sales Intelligence access section by section: Panel, Assistants, Folders and Meetings can each be read or written, and Settings can be written. Workspace admins get every section. If a custom role used Sales Intelligence before, an admin needs to tick its new sections under Account Settings, Workspaces, Roles so its members keep the menu.

A voice agent's Post-Call Analysis tab has After-call destinations. Add a row to a Google Sheet with the data you choose, create or update a HubSpot contact, update its linked deal or add a note, or run a Dapta flow with the call outcome. An agent can have up to 25 destinations, each one can be paused, and Test with last call sends the latest call to check the setup.

Voice agents can run on GPT 6.1 Sol, GPT 6 Luna and GPT 5.6 Terra, listed under LLM Model after Dapta Auto. Sol and Terra need a paid plan. On these three models, Reasoning effort goes from None or Low up to Maximum: more effort helps with hard reasoning but adds latency, and High and above warn that the caller may hear seconds of silence. Phone calls use the setting; web calls don't.

A voice agent can check availability and book meetings in Dapta Calendars during inbound, outbound and web calls. Add the Dapta Calendars Scheduling action, pick a meeting or create one with a name and a 15, 30, 45 or 60 minute duration, set the default time zone and add instructions if you want. The agent offers only times the calendar has free, in the caller's time zone.

In a voice sequence's WhatsApp fallback, the Auto-map button fills the template variables and the conversation variables with matching contact fields in one click. It uses AI and falls back to matching by name if the AI can't answer. It works when you create the sequence and later in its settings. Numbered variables, card media and button reply IDs stay for you to map.

The consumption chart on Home has a Workspace and Account switch. It opens on the current workspace's credit use, and one click shows the whole account's. Before, Home showed the workspace's use only when the workspace had a credit limit, and the whole account's otherwise.

Each contact in the CRM has an actions menu, on its row and in its panel. Call starts a call right away with the voice agent you pick, after you fill in the agent's variables. Send template opens the WhatsApp template sender for that contact, and Add to campaign adds the contact to a voice sequence. The actions need permission to edit the CRM.

The Inbox status filter takes several statuses at once and opens on open conversations, and clearing it shows every conversation. Replies you write by hand on WhatsApp can run to 4,096 characters instead of failing above 1,500. Long messages, yours and the AI's, arrive in order as consecutive messages without numbering such as (1/3).

A text agent's human handover tool now sets what the customer receives when the chat passes to a person: nothing, a fixed message you write, or a short notice the AI writes knowing whether an advisor was assigned. The AI stops answering as soon as it hands over, and when no advisor is available it still stops and sends your unavailability message. Existing handover tools start silent until you choose a message.

The Dapta MCP server accepts your Dapta login as well as API keys. Add https://mcp.dapta.ai/mcp as a connector in Claude or another assistant that supports sign-in, log in with your Dapta account, and it works as you in the workspaces where you are an owner or admin, picking the workspace per request. Each person gets their own rate limit and audit trail, and API key connections keep working as before.

Meta charges for the replies sent from WhatsApp numbers connected through Meta from 1 October 2026. When a number's WhatsApp Business account has no payment method in Meta, the agent stops trying to reply there, and the numbers screen marks replies, templates and sequences as blocked with a link to add the payment method in Meta. Replies resume on their own once Meta reports the payment method.

A Sales Intelligence assistant can be personal or shared with the workspace, run only on meetings whose title contains your keywords, and deliver its results to a webhook or a Dapta flow. A message template shapes what the assistant posts to Slack, the meeting's own details can go into any payload, and a guided tour walks you through setting up your first assistant.

Every text agent workspace can connect a WhatsApp number through Meta from Channels, either a new number or one already in the WhatsApp Business app. The redesigned WhatsApp numbers screen shows the agent that answers each number, whether it can reply and send business-initiated messages, and one fix for whatever is wrong. Photos, voice notes, videos and documents customers send reach the conversation, and Inbox replies go out on these numbers.

Wherever you pick a WhatsApp template, from the Inbox and text campaigns to the voice campaign WhatsApp fallback and a voice agent's Send WhatsApp template function, the list now holds only the templates the chosen line can send, including those saved in folders. Blank variables are caught before sending, a campaign can't start with a template its line can't use, and send errors name the template or variable that failed.

The WhatsApp template editor puts the form beside a live WhatsApp preview, in Settings, Composer and WhatsApp lines sections. Type / to insert a custom field, format text, order Reply, Link and Phone buttons and build carousel cards, and Submit for approval lists anything still missing. The templates table shows each template's status on every WhatsApp account that sends it, explains why a line can't send, and filters on every column.

Any {{name-transfer}} variable used in a transfer's custom SIP headers is now filled in by the agent at the moment it transfers the call, following an optional instruction line in the prompt, with no added delay. Test calls no longer ask you for these values. Agents that use {{call_summary_transfer}} need it renamed to {{call_summary-transfer}} to keep sending the call summary.

The voice campaign WhatsApp fallback and a voice agent's Send WhatsApp template function have a Conversation variables section listing the variables the receiving text agent declares. In the fallback each one takes a contact field; in the function, a call variable such as a policy number, or fixed text. The WhatsApp agent then knows the context when the customer replies.

The Dapta MCP server can build a Dapta database from scratch: create the database, create its tables, and add, update, upsert or delete rows, each write previewed before it is committed. It can also archive a flow or a folder, lays out the nodes of the flows it creates so they no longer open as a stack, and links knowledge bases to a voice agent, refusing one that is not ready.

Sales Intelligence settings list every connected calendar, and for each one you choose the name and image the meeting bot shows when it joins, and whether the meeting summary is emailed to the host, to the participants or to both. If Google's calendar permission box is left unchecked while connecting, a guide shows how to fix it, and a recap still being written says so instead of looking empty.

An agentic warm transfer reaches a person more often. Its screening agent says hello when the destination line stays quiet, so someone who answered in silence is not lost, and it waits through hold queues, hold music and surveys instead of taking them for a voicemail. The wait for a decision now goes up to 300 seconds, and new transfers start at 180.

Recording links that leave Dapta now follow each voice agent's Public call link switch. In the post-call webhook, call notification emails and an agent's call-log export, a public agent's recording opens at files.dapta.ai and plays straight away, while a private agent's opens the call inside Dapta after sign-in. The recording's raw storage address is no longer sent out.

A voice agent's tool results and transfer errors now follow its language: Spanish for an agent set to Spanish, and English for every other setting, including an agent with no language set. An English agent no longer ends up with Spanish failure text in its transcript or in the context it reasons over. Spoken replies and tool descriptions were already localised.

A node's credential attribute can take a runtime expression instead of a credential picked from the list, so one flow can resolve a different token per call, from a vault variable or by choosing between two agents' keys. The Option and Variable tabs both work on credential attributes, and switching back to Option restores the picker instead of trapping the attribute in variable mode.

Voice agents can run on two more models, GPT 5.6 Luna and Gemini 3.8 Flash. Choose one in the agent's LLM Model setting. Agents you have already built keep the model they are on until you change it, so nothing switches by itself.

Sales Intelligence shows what your connected calendars cost. A banner carries the credits used and left, a calendar beyond your plan asks to be upgraded or topped up before it connects, and a suspended or low-credit workspace says so rather than failing quietly. Calendars still inside their first free month show the date billing starts, the number of paid calendars and the monthly credit total.

A message template whose header carries an image, video or document can now point at a public URL instead of uploading the file. The template form offers Upload file or Use public URL, and the URL has to be a public https address; WhatsApp fetches it when the template is submitted. Switching back to uploading cancels an upload in progress and keeps the URL you typed.

Uploading a knowledge source that cannot be read now fails immediately and says why, instead of sitting in processing indefinitely. Each rejected file raises its own message naming the file and the reason: a damaged or incomplete PDF, or a format not supported yet. When only some files in a batch are rejected the rest carry on, and the source is badged so a partial upload is visible rather than silent.

Building a sequence that needs a WhatsApp channel or an approved template no longer dead-ends when you do not have one. The step shows what is missing and offers to take you straight to connecting a channel or creating a template. The prompts appear only when the prerequisite is actually absent, in English and Spanish.

The Dapta MCP server serves 130 tools instead of 92. The new ones cover flows, voice and text agents, sequences, calls, conversations, phone numbers and Dapta databases, including editing a flow by operations, comparing and rolling back flow versions, searching inside flow bodies, and testing a text agent without sending a WhatsApp message. They appear on their own in a client already connected to the server.

The AI Notetaker writes a title for each meeting from its transcript, instead of keeping whatever title the recording arrived with. The title comes from what the meeting was actually about, so a list of recordings reads as a list of topics rather than a column of identical calendar entries and generic recording names.

Phone numbers that failed verification can be removed from the phone numbers page without asking support. A delete button sits on each failed number and a confirmation dialog asks before it goes, in English and Spanish. Numbers that verified normally are untouched and keep their usual actions.

Once an account is scheduled for cancellation, a payment that lands afterwards leaves that schedule in place. Before, an invoice paid during the notice period could set the account status back as though the cancellation had never been requested. Accounts with no cancellation scheduled are unaffected and still update on payment as before.

Each Voice Agent has a Public call link switch in its call settings. Turned off, that agent's call pages and recordings open only for signed-in members of your workspace and anyone else is told the call is private. Turned on, the link works for whoever holds it, and it travels in the post-call webhook and in notifications.

The Text Agents inbox has an Awaiting first reply status filter for conversations that have opened but not yet been answered, with a banner and a tooltip explaining the state. These conversations also appear by default in chat history and in exports, so one that opened and then stalled is no longer missing from the record.

A daily email summarises the day's calls for your workspace, so call activity can be read without signing in to the platform. It covers the calls your voice agents handled that day and arrives once a day.

Call recordings in the Voice Agents call log play through Dapta itself, and the address they are fetched from is no longer shown in the table or the call detail. A loader appears while a recording starts, and a download button hands the file to your browser's own download manager, with its progress bar, instead of waiting for the whole recording to transfer first.

A text sequence no longer needs a stop date. Leave the end date empty in the creation wizard and the sequence stays active indefinitely, which suits one fed by a webhook that would otherwise be rebuilt and re-linked every time it finished. An existing sequence's end date can be cleared from its detail page. Setting an end date works exactly as before.

Encoded SIP header values can now arrive as readable Voice Agent variables and be decoded when passed on in a transfer. On an inbound SIP call a custom header is available as {{sip_header_<name>}}, and a User-to-User header or one declaring hex or base64 encoding also gets a readable {{sip_header_<name>_decoded}}. In Agent Actions, a transfer destination header can apply |decode to a value you know is encoded.

In a text agent's Meta Ad Conversions panel the separate on/off switch is gone: connect a Meta credential and choose the event that counts as a conversion, and conversions run. The credential can be added from the panel itself rather than only from Settings. Clearing the event turns conversions off and keeps the rule you wrote, and removing the credential asks first, since it is shared across the workspace.

An AI follow-up on WhatsApp can carry one attachment with a caption instead of text only, where the conversation's channel supports outbound media. Images, audio, video and documents are allowed, each checked against that channel's own type and size limits before it goes. A follow-up is attempted once, and a cancellation or a human taking the conversation over stops a send that is no longer wanted.

A voice agent's context has an optional Timezone field below Purpose, with a searchable, clearable list of timezones and labels in English and Spanish. Setting one tells the agent which local time to work from. Leaving it empty keeps the behaviour agents have today, so agents created before this are unchanged.

In Flow Studio, the Executions history marked every node of a version 1 run with a green check even when the node had returned an error, and expanding a failed node in the live Output panel showed the data going into it rather than what it returned. Each node's state is now read from what the run actually returned, so runs already in your history read correctly too, and a failed node shows its error message.

The Database node in Flow Studio accepts several statements separated by semicolons in one call. A batch runs when every statement is the same type - INSERT, UPDATE or DELETE - against the same table, every UPDATE and DELETE carries a WHERE clause, and there are at most 200 of them. The batch runs as one transaction, so if any statement fails none of them is applied. Single and parameterized queries work exactly as before.

Text agents can move a contact along your CRM pipeline. Connect GoHighLevel or HubSpot in Settings > OAuth, then open a text agent's Conversation Analysis tab, turn on CRM Pipeline, pick the connection and the pipeline, and choose which stages the agent may use. When a conversation closes, the post-conversation analysis decides whether the contact moves, at most once per conversation, and can add it to the pipeline if it is not there yet.

Dapti can now set up an outbound calling sequence for you. Tell it what you want and it takes you through the agent, the start and end dates, the retry rules, the calling days and time window, and the contacts, which you upload as a file or pull from Google Sheets. Nothing dials until you confirm the start. You can also ask Dapti to list, edit, start, stop or report on your sequences.

Voice and text campaigns are called Sequences throughout the product, in English and Spanish: the sidebar, the creation wizards, the detail and log screens, section titles and descriptions, toasts and confirmation dialogs, and the Dapta Academy videos. Only the wording changed - your sequences, their settings and their results work exactly as before.

Adding a contact that is already active in a sequence now leaves it untouched and tells you so, instead of resetting its attempts and dialing it again past the retry limits you set. Re-adding a contact that already finished resets its attempts on purpose and says so. Adding several at once reports how many were added, how many re-added and how many were already active.

Picking a template in the Dapti onboarding wizard now opens the agent's editor beside the chat and shows the build: the template's instructions type themselves into the editor, the voice is chosen in Settings, and the post-call variables are added one by one in Call Analysis. When the tour ends, the test call appears so you can talk to the agent. On screens under 800 pixels the agent is created at once and the test call appears straight away.

Agentic Warm Transfer can now be chosen on voice agents that use a SIP trunk caller ID, and any voice agent can act as the transfer agent, not only one made by the Create Transfer Agent wizard. In the call detail the briefing with the destination reads as a conversation, each tool call shows its name and how long it took, and an option records the briefing so you can play it back. Rolling out workspace by workspace.

A text agent sequence now shows what happened to each fallback call. The conversations table has a Call fallback column with that contact's state - scheduled, dispatching, called, answered, not called or failed - and the sequence metrics page adds a widget with the totals. Contacts whose fallback failed or was never placed get a Retry action that only re-dials; the WhatsApp message is never sent again.

Sending protects the line on its own. When Meta refuses a destination, that destination is blocked from receiving further messages, so nothing keeps being sent to it. Sends also fall back to the line's configured daily limit when no override is set, instead of failing. Both run on every send, with nothing to switch on and no setup step.

Text agent logs have an AI Messages column next to Messages, in both the conversations and the sessions tables. It counts the messages the agent itself sent in that conversation, so a chat where the agent did most of the talking is easy to tell apart from one where the customer wrote many times and the agent answered only rarely.

A text agent's Conversation behavior settings ask for Close after no reply for as separate Hours and Minutes fields instead of one number of minutes, with the 10-minute minimum. The three switches - Wait a moment before replying, Reply in several short messages and Show a typing indicator while replying - each sit on their own row with their description. Agents created from now on start with the last two on; existing agents keep their settings.

The Dapti header has a credits button next to History. It shows the credits used against your limit, or Unlimited, and keeps itself up to date while you work. Open it for your plan name, which links straight to Billing, the used and limit figures, a bar showing how far through the allowance you are, and an Upgrade plan button.

When a flow keeps failing with the same error, Flow Studio no longer emails its owner on every failure. The first failure of a given flow and error code is emailed straight away, further failures with that same code are held for the next two hours, and only after that can another email go out. A different error on the same flow still emails at once, and there is nothing to turn on.

Two tools are live on the Dapta MCP server and show up on their own if your client is already connected. get_flow_execution_logs lists a flow's runs newest first, filtered by status and time window, and hands you a run's call_id. get_flow_execution_detail then opens that run and returns its node-by-node timeline: each node's name, type, status, duration, input, output and the error that stopped it.

The Dapti message box now reads Ask anything… Type @ to reference an agent or flow. Type @ and a picker opens with tabs for voice agents, text agents and flows plus a search box. Pick one and it is tagged in your message and highlighted in the chat, and its details go to Dapti as context, so you can ask about that agent or flow without describing it first.

In Billing, the monthly Pro card's button now reads Get a 7-Day Trial and an Or Purchase Now link sits under it. Press the link and checkout opens with no trial attached, so your card is charged the Pro price straight away and the plan starts there and then. The trial button is unchanged for anyone who would rather have the 7 days first.

When you create a text agent campaign, choose A/B test as the template strategy and add 2 to 5 template variants with an allocation that adds up to 100%. Each new contact is assigned one variant. After launch you can reallocate the weights for contacts who haven't been assigned yet, and the campaign's metrics show delivery and engagement results by variant.

Voice agents have a new action, Send WhatsApp template. Add it from the agent's actions, pick one of your WhatsApp channels and an approved template, fill in its variables and set the number to send to, which defaults to the number being called. On outbound calls the agent sends the template while the call is still going. Without a channel or template, the action links you to create one.

Press Digit (IVR Navigation) is now in every voice agent's actions menu. Add the action, give it a name, optional instructions and a delay in milliseconds, then save. During a call, when the agent reaches an automated phone menu, it presses the keypad digits it needs to move through the menu options.

A Tasks button with a checklist icon now sits in the top bar, with a badge counting pending tasks. It opens the tasks your Dapta account team assigned to you, split into To Do and Done, with due dates and overdue labels; mark each one done or back to to do. New assignments and upcoming due dates also arrive as Task notifications in the notifications bell.

During the Pro plan free trial, the Pro plan card in Billing › Plans shows Trial active and a Get pro plan now button. Press it and a confirmation shows the card on file by its last four digits. Choose Pay Now to charge that card and activate your Pro plan right away, on the same subscription, or Change method to use a different card first.

Deleting in Flow Studio now archives instead of erasing. The confirmation counts what goes with a folder, such as 3 folders and 16 flows, and warns that their endpoints stop responding. Open Trash from the flows list to see everything archived in the workspace and restore it in one click for 30 days. A folder comes back with its contents, and each item says whether it returns to its original folder or the top level.

MCP Access now holds up to 10 active API keys per workspace instead of one, so each AI client, tool or teammate can use its own key. Creating a key asks for a name first. The list shows each key's name, prefix, creation date and last use, and lets you rename or revoke keys one by one without breaking the others. A key's full value is still shown only once, when it's created.

The Flow Studio action that sends a WhatsApp template has an optional Send at field. Choose a date and time between 1 minute and 90 days ahead with the date picker, or switch to Variable to use a date that comes from earlier in your flow, and the template is sent at that time instead of when the step runs.

In the Inbox, open the + menu in the reply box and choose Add internal note to write a note on the conversation that is never sent to the customer. Notes show in the chat as Internal note cards with author and time, and can be edited or deleted. By default the AI agent reads each note as guidance when replying; set a note to Hidden from agent to keep it for your team only.

Inbox conversations get a two-line header with tags on their own line, and the tag menu lists every tag on open and searches contact and session tags together. A Context drawer shows the conversation's status and variables, editable inline, plus a contact note the agent can read. Filters show conversation counts, applied filters appear as removable chips, and a + button starts a WhatsApp conversation from a template.

Text agents have a new switch in Conversation Behavior settings, Show a typing indicator while replying. When it's on, WhatsApp contacts see the writing… bubble while the agent works out its reply, instead of silence. Turning it on also marks the contact's messages as read, including earlier unread ones, because WhatsApp sends both together. The switch is off by default and set per agent.

In a Brain's detail page, turn on Agent can send this file for any uploaded file, or select several and choose Let agent send, then add a note on when the agent should send it. Text agents using that Brain can then attach those files in WhatsApp conversations, picking the file by its note. Sources built from web pages or text store no file, so they can't be sent. Turn a file off with Stop sending.

The Templates page for text agents has folders. Create folders and subfolders, rename or delete them, move templates between them, and search templates and folders together, with filters for category and status and a Last time used column. Files you upload as a template header are saved in Data Studio > Template Files, so you can reuse them later even if you remove them from the template.

Custom workspace roles now cover more of Dapta. In Account Settings > Workspaces, a workspace's Roles tab lets you set View, Write or None for Dapti, Sales Intelligence and Data Studio (Brains, Tables, Template images), and for Teams, Tags, voice Channels and SIP trunks. Members only see the menu items, pages and buttons their role allows, and the Users tab shows each member's role.

AI Notetaker has a redesigned meeting page. Next to the recording you get the recap, which you can copy by section, with action items you can mark as done; a transcript you can search and copy; comments tied to moments of the recording; and a Dapti tab to ask about the meeting. Folders add notes, stars and a chat across their meetings, and the dashboard shows recorded time and the last 7 days of activity.

In AI Notetaker settings, the Bot Avatar options for each connected calendar include your own banner. Upload a JPEG or PNG image (1280 x 720 px recommended, up to 1.3 MB; square logos are cropped to 16:9), select it, and the meeting bot uses that image when it joins your meetings. You can remove the banner and pick one of the preset avatars again.

Text agent settings have two more webhooks in the Webhooks section. Tag assigned runs when a tag is added to a conversation, whether a person or the agent adds it. Variable changed runs when one or more conversation variables change and includes their previous and current values. Both are signed the same way as the agent's other webhooks, so your flow or system can react as soon as a conversation is classified or updated.

The Inbox filter panel has a Date range section. Pick a From date, a To date or both, and the list shows the conversations with messages in that period. A footer under the conversation list now shows how many conversations match your current filters, with or without filters applied. It reads "At least N conversations" for a moment while the exact count loads.

A voice agent can pass a short summary of the call to whoever receives a transfer. In the Transfer Call action, add a custom SIP header and set its value to {{call_summary_transfer}}. When the agent transfers, it writes up to two sentences with the reason for the call and where it stands, and sends them in that header to the receiving phone system. Rolling out workspace by workspace.

Template Images in Data Studio is now Template Files. Along with JPG and PNG images, you can upload MP4 and 3GP videos and PDF, TXT, Word, Excel and PowerPoint documents, keep them in folders and copy each file's URL to use as the media header of a WhatsApp template.

When you import a contact file into a voice or text agent campaign, valid numbers from any country are accepted, whatever the campaign's time zone. Mexican numbers in the old +521 format and country codes written twice are corrected automatically, a number repeated in the same file updates its contact, and the import result says how many numbers were read as another country. Error messages name the real cause of each failed row.

The Transfer Call action for voice agents has a Restrict to business hours switch. Turn it on and pick the days, a start and end time and a timezone. Outside those hours the agent doesn't transfer the call, with no prompt rules needed. The switch is off by default, so existing agents behave the same. Rolling out workspace by workspace.

Flow Studio has new OpenAI, Claude and Google Gemini AI nodes that share one form. Pick a model from a list rated for reasoning, speed and price, write the prompt with variables from earlier steps, and attach up to 5 files by link, such as a Google Drive file (PDF, Word, PNG, JPG, WebP or GIF, 15 MB each). Run the node on Dapta AI or connect your own API key. Older AI nodes show a notice to switch.

Text agent settings have a Conversation behavior section. Set how long a conversation stays open without a reply, from 10 to 1,440 minutes. Turn on Wait a moment before replying so the agent pauses briefly before sending, or Reply in several short messages to split an answer into up to 3 texts, each counted as a message. In Webhooks, Message not delivered runs a flow when a message doesn't reach your customer.

When you add a Custom Action to a voice agent, a Store fields as variables section lets you pick fields from the flow's response, such as response.tracking, and give each one a variable name. During the call, the agent saves those values as variables it can use in later steps, for example in a transfer message.

Tables in Data Studio open full screen with Records and Settings tabs. Edit a record in a side panel, import or export CSV, and manage fields with friendly or advanced types and drag-and-drop order. On a DaptaDB database hosted by Dapta, field changes update the real columns; for a database you connected, Sync fields adds new columns or removes missing ones without changing it. New creates a DaptaDB database or connects a source.

After a few exchanges with Dapti, a short card above the message box asks "How is Dapti doing this session?". Choose Bad, Fine or Good, by click or with the 1–3 keys; after Bad or Fine you can add an optional comment. The card shows at most once per conversation, waits seven days before asking again, and never appears during a test call or a text agent trial.

Voice campaigns have a Smart callbacks option in Advanced Settings. When a contact asks during a call to be called back at a specific time, the campaign schedules that call automatically. Set how many days ahead a request can be and a default delay for when no time is given. A Scheduled call column and the contact panel show each pending callback or re-dial.

When you start the 7-day free trial of the Pro plan, checkout takes you to Dapti, which greets you with a welcome: your trial has started, and do you already have an agent or want to build one from scratch? Reply and Dapti guides you from there, asking whether you need a voice or a text agent. If you leave before replying, Dapti keeps that context for when you come back.

Inbox filters have a Meta Ad section listing the Click-to-WhatsApp ads your conversations started from, each with its ad ID and number of conversations. Select one or more ads to narrow the list, and clear any of them from the pills above it. Click the Meta Ad badge on a conversation to copy its source, ad ID and click ID.

When Dapti creates a voice agent for you, the new agent's editor opens beside the chat and you watch the build: Dapti writes the agent's instructions live, sets the voice and call settings and adds the post-call variables, then tells you the agent is ready to try. Agents in Spanish get their post-call variables named and described in Spanish.

The Transfer Call action for voice agents has a Personalize transfer message switch. Turn it on and describe the announcement in the action's instructions, for example that the caller is being connected with a specialist. Before the transfer starts, the agent says your message instead of the standard hold message. Rolling out workspace by workspace.

Open conversations in Inbox have a Tags menu at the top of the chat. Search your workspace tags or type a new name to create one, then add it By session, for this conversation only, or By contact, so it stays with the contact across conversations. Session tags clear when the session closes. Remove a tag from its chip, and use the Tags filter to find tagged conversations.

The Dapta landing page has a prompt box: describe the agent you want and press send. A sign-in window opens where you continue with Google or Microsoft, or with your email and a one-time code. You land in Dapta with a Dapti conversation already started from your prompt. New accounts go through the usual onboarding first, and Dapti picks up the prompt right after.

In the row of actions under each Dapti reply, a new Export button sits next to Copy. Choose Export as PDF or Export as DOCX and the reply downloads as a file. The PDF keeps headings, lists, tables and code blocks, shows the Dapta logo, your workspace name and the date at the top of each page, links any web sources Dapti used, and numbers its pages.

The Tables page in Data Studio lists your tables as a tree: each data source, then its schemas, then the tables inside, with a count per group and each item's type (table or view) and last edit. Search from the header, add one with New, and open a table to view and manage its records, with a Back button to return. Tables not connected to a data source are grouped on their own.

In a voice campaign's Advanced Settings, Auto redial immediately calls a contact back when they don't answer, only on the campaign's first scheduled attempt. Set Retry after in minutes and Maximum re-dials, up to 3. Re-dials no longer count toward the campaign's max attempts, and the contact drawer and the contact's campaign list in CRM show re-dials as their own counter.

Workspace owners can remove people from a workspace. In account settings, open the workspace and go to the Users tab, where every member except the owner has a delete icon. After you confirm, the person loses access to the workspace. To bring them back, invite them to the workspace again.

Dapti's memory now has two parts. In account settings, the Dapti page has a My Memory tab with your name, contact details and preferences, private to you, and a Workspace tab with shared company context, such as your industry, agents and integrations, that Dapti uses no matter who is chatting. Review, edit or remove what either one holds and save. New accounts start with both filled in from their onboarding answers.

The Transfer to a human action in a text agent now asks who to transfer to. Pick Member to send chats to one person, and write the message contacts see when that person isn't available. Pick Team to use a transfer team and choose how chats are assigned: Fixed goes through the members in the same order, and Balanced gives each chat to whoever has the fewest active chats.

Meetings in AI Notetaker can carry tags. Add them from a meeting's detail view or from the meetings list, create a new tag on the spot and pick its color, and remove one from its chip. Filter the list by tags, matching any or all of them, and select several meetings to tag them at once. A Tags column shows each meeting's tags.

Ask Dapti for a report, a PDF or a Word document and it builds one from the analysis in your chat, with charts when they help. A download card with a PDF or DOCX icon appears in the thread. Open it to download the PDF, the DOCX, or both, depending on what you asked for.

When a contact starts a WhatsApp conversation with a text agent by tapping a Meta Click-to-WhatsApp ad, a Meta Ad badge marks it. It appears on the message, on the chat in the Inbox list and header, and in Text Logs sessions and conversations. Hover over it to see the ad's headline and text, the source, the ad ID and the click ID. Conversations that didn't start from an ad show no badge.

Every text agent and every text campaign has a Metrics tab. Pick a date range to see credits used, contacts, responses, connection rate, successful responses and average messages per response, along with a contact funnel and a responses-per-day chart. Test conversations from the playground are left out, so the numbers reflect real contacts.

The Create New Agent button in Voice Agents and Text Agents opens two choices. "Create with Dapti" takes you to Dapti with the request to build a new voice or text agent already sent, so you shape the agent in conversation. "Create from scratch" opens the step-by-step setup wizard.

If you start upgrading from the Free plan and leave checkout without paying, a reminder appears in the top right corner when you return to Dapta. "Complete payment" reopens checkout for the same plan in one click, "I need help" puts you in touch with support, and you can close the reminder at any time. It disappears once your plan is active.

MCP Access, in Account Settings, connects your AI assistant to a Dapta workspace. Create an API key for the workspace, then follow the tab for your client: Claude Desktop gets a ready-to-upload plugin with your key inside, and Claude Code, Codex and Cursor get setup steps to copy. Your assistant can then work with your voice and text agents and Flow Studio right from a chat.

docs.dapta.ai has a new section on connecting your telephony (SIP trunk) to Dapta, in English and Spanish. It covers connecting a Dapta-managed number, bringing your own SIP trunk from your provider, and custom telephony for enterprises and large companies, with screenshots of each setup step in Dapta.

A three-dot menu at the top of each Inbox conversation holds Close conversation, Block user and Unblock user. After you confirm, a blocked contact can no longer message that channel, and unblocking lets them write again. Messages written by AI show the name of the text agent that sent them, so after a transfer between agents you can tell which one answered.

Dapti answers questions about the HubSpot account connected to your workspace: its lists, properties and owners, and the contacts, companies or deals that match what you ask. It only reads HubSpot data and never changes it, and it can use what it finds to build a Flow Studio flow, such as one that calls the leads in a list. If HubSpot isn't connected yet, Dapti explains how to connect it.

Voice campaigns have a "Re-dial the contact if they don't answer" switch in Advanced Settings. Set how many minutes to wait before calling again (0 re-dials right away, up to 7 days) and, if you like, a maximum of 1 to 3 re-dials. When a call goes unanswered, the campaign schedules another call to the same contact, within its max attempts and calling schedule.

Importing a CSV into a Data Studio table shows one row per file column, each with a searchable field picker and a Skip this column option. Columns are matched to fields by name even when case, accents or separators differ, the table's ID field is preselected for updates, and the Upload button shows how many rows are going in. Values with commas inside quotes stay in their own column.

WhatsApp campaigns for text agents have a Call fallback section. Turn on "Call unanswered contacts", pick a voice agent and set how long to wait before calling, from 1 to 24 hours. Each contact who hasn't replied to the template by then gets one automated call, within the campaign's schedule windows and before its end date. You can also map the voice agent's variables to contact fields.

Voice campaigns have a WhatsApp fallback in Advanced Settings. Turn it on, pick the WhatsApp number and an approved template, and map the template variables to contact fields. When a contact runs out of call attempts without connecting, Dapta sends that template as a last touch, or sends it after the attempt number you choose.

When you create a WhatsApp campaign for a text agent, the contacts step offers four sources: upload a CSV or XLSX file, or sync contacts live from a Google Sheet, a HubSpot list or a webhook URL. Variable mapping uses the columns of the connected source. The campaign's Connections tab lists its live sources with Sync now, Pause, Resume and Disconnect, and disconnecting keeps contacts already imported.

The Billing plans page, and the upgrade window on the Free plan, have a Monthly / Yearly toggle. On Yearly, Pro and Scale Up show their annual price and how much you save, and credits still arrive every month of the annual subscription. Monthly subscribers can move to a yearly plan at any time, and annual subscribers can schedule a switch back to monthly that takes effect when the annual period ends.

Conversation Tags, in the Text Agents menu, lets you create tags with a name, a color, instructions for when the AI should apply them and a scope: permanent, or only for the current chat session. In each text agent's settings, choose up to 30 tags it can use. The agent applies them to conversations following your instructions, and Inbox shows the tags on each chat and filters by them.

Email notifications for voice agents have a Subject field. Write the subject line the recipient sees after each call, and insert input variables or post-call analysis variables with {{variable}}, for example "Call summary for {{fullName}}". Leave it empty to keep the default subject. Test emails show the subject with sample values filled in.

When Dapti answers with numbers, such as calls per agent or results over time, it draws a bar, line, area, pie or scatter chart right in its reply. When Dapti opens one of your voice or text agents in its side panel, you get a compact editor with Test Agent, Settings and Instruction Set tabs, so you can test and adjust the agent without leaving the conversation. Drag the panel's edge to resize it.

On phone calls, voice agents fill in {{variables}} inside Call Transfer and Agent Transfer actions: in the action description, in the prompt used to pick the transfer number, and in the message the agent says during an agent transfer. Each call uses the contact's values, and built-in variables such as the current time, call direction and phone numbers work too. Variables with no value stay as written.

Optimize your calls by automatically dividing your contact list among different voice agents with customizable percentage weights. Measure the actual performance of each strategy from a side-by-side comparative dashboard that analyzes connection rates, costs, and goals achieved to identify the winning agent and maximize your conversions.

Voice agents can now speak with emotion. Choose a voice with an Emotionality rating in Select Voice, and the agent sets its tone sentence by sentence (content, excited, sad, angry or scared) to match the conversation, in English and Spanish. You don't need to change your prompt: the emotion guidance is added for you, and each emotion used appears as a tag in the call transcript. Rolling out workspace by workspace.

A new "Test on WhatsApp with a QR" option in the text agent test chat lets you try an agent on WhatsApp from your phone. Scan the QR code or open the link, send the pre-filled message with its test code (valid for 24 hours), and you are chatting with your agent on a shared Dapta test number.

Voice agents have a Pronunciation section in Speech Settings. Add a word as it is written and how it should sound, press play to hear it in the agent's voice, then save. On calls, the agent says your version instead. It works well for brand names, people's names, acronyms and foreign terms. A rule matches the text wherever it appears, even inside longer words, so use distinctive words. Rolling out workspace by workspace.

When you set up a WhatsApp campaign for a text agent, the variable mapping step now lets you link columns from your contact file to the conversation variables your agent defines, next to the template variables. Each contact's values travel with their conversation, so the agent can use them during the chat. The section is optional and only appears when the agent has conversation variables.

Text agent settings include a model selector. Choose between premium and lower-cost AI models, and each option shows how many credits it uses per message, so you can balance quality and spend agent by agent. Lower-cost models can't process images or audio.

This update allows users to fuel voice campaigns through a live connection with Google Sheets, moving beyond static CSV uploads. After a quick Google account link and automatic column mapping, the system checks the sheet every 15 minutes to automatically synchronize and queue new leads, paving the way for upcoming webhook and CRM integrations.

This new feature allows users to automate recurring operational actions and continuous workflows within Dapti based on set periodicities. Designed to build daily product habits and reduce manual tracking, the system automates follow-ups, with upcoming integrations planned for Gmail and Google Calendar to seamlessly sync Dapti-generated tasks and reminders directly into the user's daily workflow.

This update eliminates abrupt hang-ups by integrating a guaranteed farewell message directly into the End Call tool, managing transitions seamlessly via a new configuration toggle. Additionally, the platform now automatically handles unintelligible audio from noisy environments, prompting users to repeat themselves instead of falling silent, ensuring a natural and continuous conversational flow.

This feature introduces Model Context Protocol (MCP) support to connect Dapta directly with external AI systems like Claude. It allows users to programmatically manage API keys, install plugins, and build or modify agents through cloud platforms, enabling seamless agent management and creation from external AI environments.

To deliver a more fluid and natural user experience, this new toggle allows agents to be interrupted while processing and delivering text responses from tools. The toggle is off by default and appears on most voice agent actions except End Call and Press Digit, so you choose which actions callers can interrupt. Rolling out workspace by workspace.

The Call Analysis tab now features an "Improve with AI" option for the Call Objective field. Designed to assist users in defining clear goals, this tool analyzes the agent's prompt and context to automatically generate an objective suggestion in the matching language. Moving forward, the product team is refining the AI logic to ensure it suggests a single, concise objective rather than multiple combined goals, keeping post-call success metrics and "cost per result" tracking highly accurate and easy to evaluate.

To sustain the speed and computing power your agents require, Dapti is transitioning to a usage-based credit model. The Free plan now includes 2,000 monthly credits for testing, while the Pro plan ($99/mo) comes packed with 100,000 monthly credits to scale your operations without friction. With an average message costing just 150 credits (~$0.15), your Pro allocation is built to go a long way.

This feature introduces automated email alerts triggered by specific conditions defined during post-call analysis, utilizing boolean or selector variables. The system is designed to streamline follow-ups based on call outcomes, with future iterations planned to include advanced variable selection, dynamic inputs, and Slack integration.

This feature allows users to view, track, and restore historical versions of their text agents with ease. Version histories load dynamically and include detailed timestamps for every modification, giving users full control to revert changes and maintain conversational consistency.

Voice agents have a new transfer type, Agentic Warm Transfer. Before the caller is connected, a transfer agent talks with the destination and decides whether to bridge the call or cancel it and return the caller to the main agent. Create the transfer agent from the Transfer Call action, and its prompt and actions are set up for you. At launch it isn't available in every workspace or for agents that use a SIP trunk caller ID.

The prompt editor received a major update to streamline the creation process for non-technical users. It introduces a rich-text "Preview" tab that hides technical code, refined "Improve with AI" functionality to modify only specific sections, and Notion-style "slash" commands for easier content structuring. Additionally, updated model selection guidance now recommends model 4.1 for speed and cost-efficiency, and 5.1 for complex tool-calling tasks.

A new invoices screen was introduced to resolve a bug that limited table displays. The updated UI features smart filters, live data from Stripe, and contextual actions. It is currently in production and responsive for mobile use.

This update introduces a comprehensive quality framework that evaluates agent performance across three core pillars: Goal/Result Quality (task achievement), Conversational Quality (communication execution), and Technical Quality (latency and system performance). Designed to provide a holistic view of operations, future iterations will display these metrics along with "cost per result" directly to the user to optimize decision-making and agent efficiency.

Rebuild the "Improve with AI" functionality in Voice Agents: fix the shortening/deleting of prompts, use the Dapti engine, improve UX with "Entire Prompt" or "Only a Section" options.

This feature was created to evaluate calls from the Call Log, displaying scores and details to help iterate and quickly improve the quality of voice agents.

Improve date range filters for analysis: support presets, custom ranges, consistent formatting, and avoid interpretation errors that break reports or scheduling.

Redesign the contextual roles and permissions system in accounts and workspaces, with granular RBAC, inheritance, wildcards, impersonation, and secure multi-account access.

Update Flow Studio UI with more consistent interfaces: better tables, advanced filters, dynamic columns, and a clearer experience for campaigns and automations.

This newly built landing page provides a comprehensive overview of text agents, detailing how they function through building, training, launching, and understanding. It highlights key integrations for WhatsApp and website widgets, showcases various industry use cases, and includes customer success stories

Now Available on both the App Store and Google Play, this initial release focuses on an inbox feature that allows human operators to monitor and respond to text agent conversations. It provides a familiar WhatsApp-style interface for managing active conversations across different workspaces.

This update introduces "Feedback" and "Tools" columns to the call log table. The Tools column displays the number and specifics of functions executed during a call. The Feedback column tracks call results, including positive/negative ratings (likes/dislikes) and user comments, all of which are now filterable.

Integrated directly into the platform, this section provides centralized access to video tutorials previously hosted only on YouTube. Users can toggle between English and Spanish content organized by feature. Future plans include adding quizzes for credits and certifications for implementation partners.

A mobile application is currently operational on Android devices. This initial version specifically supports the inbox functionality for text agents. The app allows users to manage these agent interactions while on the go, expanding the platform's accessibility beyond the desktop interface.

New skills enable the AI to perform complex tasks like analyzing its own performance metrics and querying database structures. Most notably, a new skill allows Dapti to automatically build and modify functional workflows in Flow Studio based on natural language requests, including node validation.

The company is transitioning Pro plan users to a self-service onboarding model. Personalized 1-on-1 implementation sessions for new Pro plan users ended on March 13, 2026. Moving forward, Pro users will be guided by a new smart, chatbot-like interface designed to help them build agents independently.

In the first experience, when the user selects Voice Agents, we changed the first screen that appears, only explaining the difference between a template agent (quick creation) and a custom agent (advanced creation).

This new hub features a "What’s New" section for product updates and a notification area for maintenance alerts, webinars, and office hours. Managed via a back-office dashboard, it allows targeting messages by user group, country, language, or subscription plan (free, pro, scala, or prime) to improve feature awareness.

This feature integrates WhatsApp business connections with both voice and text agents. Users can connect a verified phone number to an agent through a new connections panel. Crucially, the number must undergo Meta business verification to function.

This tool extracts structured information, such as sentiment and success analysis, once a text conversation ends. Users define variables that are sent via a JSON to a flow URL for further processing, such as updating CRM fields.

Currently in beta, this feature allows users to purchase Colombian and Mexican phone numbers directly through the platform by completing a regulatory bundle. It requires submitting business or individual documentation, such as identification and address verification, for approval by Twilio.

This is an iteration of the gamification feature; we've prioritized some challenges and removed others. Each challenge awards credits but aims to guide the user to spend credits through web calls that already cost credits.

These updates focus on efficiency and safety during campaign execution. Users can now run manual campaigns immediately without waiting for scheduled times. Additionally, custom batch sizes allow for specific contact amounts per round, and phone number rotation prevents numbers from being "burnt" or flagged during active runs.

This feature replaces the old multi-step setup with a streamlined, template-based onboarding process. It offers pre-built options like lead qualification or receptionist agents, and guides users through language selection and starting their first call.

This knowledge-base feature on Text Agents allows agents to answer questions using uploaded documents like PDFs, CSVs, or plain text. Users can create new "brains" or reuse existing ones across both voice and text agents. Recent efforts focus on polishing PDF quality and fixing bugs related to saving brain configurations.

The new filters in the voice selection feature allow users to refine their search by accent, gender, age, and voice model providers. These improvements, supported by metadata, enable faster, clearer selection and better customization for clients, including the ability to restrict exclusive voices to specific customers.

This gamification feature uses an "earn while you learn" model to onboard users. Users complete sequential tasks—such as creating company contexts, making test calls, and editing variables—to earn credits. It aims to accelerate user activation and demonstrate platform value through guided interaction

The redesigned home replaces empty dashboards with clear entry points and popular use cases for voice and text agents. Designed to drive day-one activation, it includes credit usage tracking and links to documentation and tutorials. The goal is to reduce onboarding drop-off and improve week-one retention.

This update moves template management from the channel level to the organization level, allowing use across multiple lines and agents. Users can create, save drafts, and submit templates for WhatsApp approval directly within the platform. New features include template duplication and the ability to attach multimedia documents.

The "human in the loop" feature allows AI agents to transfer active conversations to human operators via an "agent transfer" action. Users define transfer triggers and can auto-assign chats to specific teams or individual agents for manual handling

The "Call Me Back" feature creates a bridge between text and voice agents via a website widget. It provides a form for end customers to request an instant call, immediately dialing and connecting them with an AI voice agent.