Changelog

If you run more than one channel, the top of your dashboard has been asking you to do arithmetic it never showed its working for. Total counted everything. The status cards next to it counted chats. On a day with two web chats and six emails you would read Total 8 beside a breakdown that added up to two, and reasonably conclude six conversations had gone unresolved. They had not. They were emails, and emails were never in that breakdown. Today's Overview now says so.

See the mix at a glance

• A volume bar above the cards. Today's total is now shown as a single bar split by channel, with a count and a percentage for each one. Web chat, WhatsApp, Instagram, Messenger, email and voice each get their own segment, in proportion.
• Readable without relying on colour. Every segment is labelled with its name and its number, so the bar reads correctly in greyscale, at a glance, or for anyone who does not distinguish the colours.
• It stays out of the way when it has nothing to add. If you only use one channel, the bar does not appear. There would be nothing to break down, and your dashboard looks exactly as it did before.

The status cards now state their scope

Conversation Status carries a badge telling you exactly how many of today's conversations it covers, for example Chats only, 2 of 8. The Total card spells out the same split underneath its number, such as 2 chats, 6 email. No more guessing which cards describe which conversations.

Email has its own row

Email conversations do not behave like chats. A chat resolves in minutes while you are watching it. An email thread can sit for days waiting on a reply, so forcing it into the same resolved and abandoned buckets was never going to tell you anything true. Email now reports on its own terms, and the three cards always add up to your email count for the day.

• Awaiting Reply. The customer sent the last message and nobody has responded yet. This is the number worth watching.
• Replied. You or your AI agent sent a reply on the thread.
• Dismissed. The draft was cleared from your SYNC inbox without a reply being sent, because none was needed.

Your numbers have not changed

Nothing was recalculated. Total still counts every channel, and Resolution Rate, Abandon Rate and Transfer Rate are still measured against your chat channels, exactly as before. This release changes what the dashboard tells you about those numbers, not the numbers themselves. The updated dashboard guide walks through every card and what it counts.

If your working day did not end at midnight UTC, your dashboard has been resetting too early. Today's Overview measured a day by the server's clock, so for a team five hours behind UTC the counters rolled over at 7pm, and for a team in Sydney they rolled over at 10am. Conversations handled in the afternoon quietly moved into tomorrow while the day was still going. That is fixed. A day now begins and ends at midnight where you are.

What this changes

• Today means today. In Today's Overview, Total, the status cards, the email row, CSAT, NPS and both response times all cover midnight to now in your own timezone. No more finishing the day looking at a dashboard that has already reset to zero.
• Daily bars line up with your days. Every daily point on the conversation charts is now your calendar day. Previously a bar labelled Monday actually held part of Sunday evening and stopped short of Monday evening, which is why the totals could look shifted.
• Agent status days match shifts. In the agent status report, time on availability is attributed to the day the shift actually happened, so an evening on duty no longer lands on the next day's row.
• Nothing to configure. Your timezone is detected automatically from your browser. There is no setting to find and no default to get wrong. If you travel, the dashboard follows you.

Your history has not been rewritten

No conversation moved and no total was recalculated. The only thing that changed is where one day is cut from the next. You may notice a daily figure shift by a few conversations around the boundary compared to what you saw before, and the new number is the accurate one. Reports built on rolling windows, such as the last 7 or 30 days, are unaffected: they never had a midnight to get wrong.

When we ship a new build, the app lets you know. Until now that prompt was all or nothing: reload straight away, or dismiss it and have no way back until you happened to refresh. If you were halfway through something, neither option was any good.

• Later actually means later. Dismissing the prompt moves it to a small marker above your name in the sidebar, where it waits indefinitely. Update whenever you reach a natural stopping point.
• You can see what changed. The prompt shows the version you are on, the version you are moving to, and a link to the release notes that opens in a new tab so you keep your place.
• It asks once. The full prompt appears a single time. After that the sidebar marker is the only reminder, so you are never interrupted twice for the same thing.
• Quiet about quiet releases. Behind-the-scenes fixes no longer announce a version you have already been told about. They simply offer the update.

Putting a voice agent on your phone line is a leap of faith: you only find out how it handles a hostile caller, a refund it should refuse, or someone fishing for another customer's details once a real person is on the line. Echo now lets you find out first. Simulations play out the hard calls and grade the agent against them, and a new settings hub puts the rules it follows in your hands.

Simulations: grade the agent before it goes live

• A test suite that starts full. Simulations opens with eight high-risk caller situations already seeded: refunds, cancellations, identity checks, escalation, prompt injection, vulnerable callers, over-promising, and privacy. Toggle any of them on or off, reword them, or write your own, up to twenty.
• Graded against a standard you cannot weaken. Every run is scored by Elysium, our proprietary guardrail engine, across policy adherence, safety and data privacy, identity verification, anti-hallucination, over-promising, escalation, and tone. Your own criteria layer on top of it, never underneath it.
• Your pass mark, your rules. Set the score a scenario has to reach to pass, and add the criteria that are specific to your business, such as always mentioning the returns window or never discussing a competitor's pricing.
• Results you can act on. A run gives you an overall pass or fail, a breakdown by group, and a card per scenario with its score and a one-line reason. Past runs stay listed, so you can watch the score move as you tighten the agent.
• No calls, no minutes. A simulation grades the agent's configuration. It does not dial anyone and it does not spend call minutes.

The loop: a failure becomes a guardrail

When a scenario fails, the run proposes the guardrail that would have closed the gap, right under the result. Select Add guardrail and it drops straight into your active rules. Re-save the agent, run again, and watch the weak spot disappear.

Guardrails, procedures, and approved actions

• Hundreds of guardrails, always on. Every Echo agent runs on a built-in safety baseline covering prompt-injection resistance, system-prompt protection, identity and account protection, data privacy, anti-hallucination, scope control, tone, abuse handling, and human handoff. It cannot be turned off, and nothing you add can weaken it.
• Your own rules on top. Write a rule, name it, and choose what happens when it trips: flag for review, escalate to a human, or block. Or add a proven one from the recommended list, grouped by safety, honesty, privacy, identity, and escalation.
• Standard operating procedures. Give the agent a step-by-step process for a specific job, such as a refund or a claim, and it follows those steps on every call.
• Approved actions. Decide what the agent may do on its own, what needs the caller's explicit confirmation, and what has to go to a person.

Documentation, with screenshots

Three new step-by-step guides shipped alongside the feature on the same day, each walking through the real screens: Echo Simulations, Echo Guardrails and Safety, and Echo Outbound Campaigns.

We rewrote the documentation from the ground up. Every guide now matches what you actually see in the product, names the exact buttons to click, and shows real screenshots from the app.

• Accurate, step-by-step guides. All sixty-plus articles were rewritten against the live product, so the steps match the screens.
• Real screenshots. How-to and integration guides are illustrated with captures from the running app, not placeholders.
• Search built in. Press Cmd or Ctrl plus K anywhere in the docs to jump straight to what you need.
• Easier to read. Syntax-highlighted code, copy buttons, callouts, and an on-page table of contents on every article.

Putting ChatFlow on your store should not require a developer. Shopify and Magento now install with a one-click connect, so the widget goes live without editing a theme or pasting a snippet.

• Shopify one-click install. Authorize once and ChatFlow adds the widget to your storefront automatically.
• Magento connect flow. Set up Magento from inside the dashboard, with a guided connect flow.
• Already supported. Wix, WordPress, and Webflow connect the same way, no code required.

It should always be clear what you can do in a workspace and why. We added a place to see your role and tightened how roles map to what is available.

• Your role, at a glance. A new card in Profile, under My Details, shows the role you hold and the organization it applies to.
• Plan-aware roles. The roles you can assign now match your plan, so the options you see are the ones you can actually use.
• Consistent page access. Pages and actions are gated by role on both the screen and the server, so the dashboard matches your permissions.
• Owner changes stay protected. Only owners and admins can change roles, and ownership changes are guarded.

Sentiment analysis grew from a read-only chart into a way to act. You can now turn a frustrated conversation into a tracked task, assign it, and get alerted before a single complaint becomes a trend.

• Escalate an issue. Promote a conversation to a tracked issue in one click when it needs more than a quick reply.
• Schedule and assign follow-ups. Set a reminder, pick an owner, and the assignee is emailed so the loop actually closes.
• Track themes by phrase. Define the complaints you care about in your own words and watch how often they come up.
• Negative-trend alerts. Admins and owners are notified when sentiment on a theme starts sliding, not after it has.

We brought every surface of ChatFlow under one design system, anchored on a single cobalt brand color. The result is calmer, more legible, and easier to scan when you are working through a queue.

• One system, every surface. The agent console, the conversation viewer, and the website now share the same colors, spacing, and motion.
• Color that carries meaning. Status tones tell you at a glance what is resolved, pending, or needs attention, instead of decoration.
• Clear sender legibility. Human agent, AI, system, and customer are now instantly distinguishable in any conversation.
• Refined loading and empty states. Skeletons and honest empty states replace blank screens and spinners throughout the app.

Until now, a scanned PDF or a photo of a document was just an image to the assistant. With OCR, ChatFlow extracts the text inside those files, so the answers your customers get can come from them too.

• Scanned PDFs and images. Upload a scan or a screenshot and the assistant reads the words inside it.
• Legacy Word files. Older .doc files (Word 97 to 2003) now upload alongside modern formats.
• Works where you already train. Add these files from the Documents tab of any chatbot, the same place as the rest of your knowledge base.

New accounts now start in a guided wizard instead of an empty dashboard. It asks what you want your assistant to do, helps you train it, and gets you to a working chatbot you can test in minutes.

• Goal-based setup. Tell us what you want to achieve and the wizard tailors the steps to match.
• Faster first answer. Train the assistant on your top questions during setup, so it is useful right away.
• A checklist that remembers you. Your progress is saved on the server, so you can leave and pick up exactly where you stopped.
• Help when you stall. Contextual prompts appear if you get stuck, instead of leaving you to figure it out alone.

Insurance and other regulated businesses can now connect a policy system to ChatFlow. The assistant verifies who it is talking to before it discloses anything, then answers account questions directly in the chat.

• Identity verification first. The assistant confirms a customer's identity across multiple factors before sharing any account detail.
• Policy lookup. Customers can ask about coverage, status, and endorsements and get an answer without waiting for an agent.
• Document access. Share policy documents securely inside the conversation.
• Balances and endorsements. Report outstanding balances and recent changes on request.

We made a broad pass over how ChatFlow handles untrusted input and sensitive data. These protections are on by default for every account, with no configuration required.

• Message encryption across channels. Conversations on the web widget, WhatsApp, Instagram, Messenger, and email are encrypted at rest.
• Prompt-injection defense. The assistant detects and refuses attempts to override its instructions or extract data it should not share.
• Attachment malware scanning. Every file is scanned on the way in and again before it reaches an agent, so malicious uploads are blocked.
• Content moderation. Unsafe training data and abusive messages are blocked, with a strikes-based policy for repeat offenders.

Echo brings ChatFlow to the phone. An AI voice agent answers from the same knowledge base your chatbot uses, holds a natural conversation, and hands off to a human agent when the caller asks or the topic needs it.

• Answer inbound calls. Echo picks up, understands the caller, and answers from your knowledge base, around the clock.
• Outbound voice campaigns. Schedule calls to follow up with customers or run reminders at scale.
• Natural voices. Choose from OpenAI Realtime and ElevenLabs Turbo voices to match your brand.
• Bring your own SIP trunk. Connect existing telephony, or provision a new number inside ChatFlow.
• Human handoff. Echo knows when to transfer to a live agent, so callers are never stuck.