Changelog

Follow up on the latest improvements and updates.

RSS

20250805_Jupiter_ArtCard_1920x1080_16x9_DZ_v20__00265___1_
OpenAI GPT-5
The newest chat models from OpenAI beat all predecessors in terms of intelligence, speed and price.
Starting today, you can
upgrade your chatbot to GPT-5 and GPT-5-mini
. Both models perform
exceptionally well
and I'm excited we can make them available to you.
We recommend testing your chatbot after the change to make sure it works as expected on the new model.
Claude, Gemini, Mistral
We just enabled the updated versions of the following flagship models:
Each of those models offers superior performance compared to their predecessors. You can compare all our available models in this guide.
Deprecation Notice
Large Language Models improve at an astonishing pace. With the newest models leaping forward, some old - once the best - models are now outdated.
To streamline our offering, make customer support faster and product easier to configure, we've decided to deprecate some of the oldest models:
  • OpenAI GPT-4o
  • OpenAI GPT-4o-mini
  • Anthropic Claude 3.7
  • Google Gemini Flash 2.0
  • Mistral Large
The above will be deprecated on November 1st
. We recommend switching to their successors before that date.
Chatwith is an EU-based company, and more than half of our customers are located in Europe.
We've decided to better
support local businesses and progressively enable more parts of our systems in Europe.
The following infrastructure changes have been made recently:
You can now choose
Mistral Large
for your chatbots. It is a flagship model from a
French AI company 🇫🇷
, with servers located in the EU.
We've enabled
EU data centers
🇪🇺 for our cloud computing. This results in faster chatbot loading for visitors from the region.
A few months ago we've migrated our
payment services
to Paddle, a UK company 🇬🇧.
We're committed to do the same to all the remaining parts of our infrastructure (eg. the database) and eventually migrate all our services to EU-based companies.
1_U5jlTiqa3SiMrS_lFeqytw
GPT 4.1
The newest chat models from OpenAI beat all predecessors in terms of intelligence, speed and price.
Starting today, you can
upgrade your chatbot to GPT 4.1 and GPT 4.1 mini
. Both models perform
exceptionally well
and I'm excited we can make them available to you.
We recommend testing your chatbot after the change to make sure it works as expected on the new model.
Claude 3.7
We now offer the updated version of Anthropic Claude, 3.7. It offers:
  • 94% accuracy in multi-turn conversations (vs 3.5's 87%)
  • 12% better logical reasoning accuracy than 3.5
  • 0.97s latency vs 3.7's 1.2s average response time
  • ... and more - all at the same price!
This is an incremental, yet obvious improvement. We automatically upgraded all customers using Anthropic Claude, with no action required or extra cost on your side.
Deprecation Notice
Large Language Models improve at an astonishing pace. With the newest models leaping forward, some old - once the best - models are now outdated.
To streamline our offering, make customer support faster and product easier to configure, we've decided to deprecate some of the oldest models:
  • GPT-3.5 Turbo
  • GPT-4
  • GPT-4 Turbo
The above will be deprecated on May 14th
. We recommend switching to either GPT-4.1 or GPT-4.1-mini before that date.
AI Chatbots are increasingly becoming a part of the funnel. They often lead to, or help close, a conversion.
If you use a chatbot as part of your marketing or sales efforts, it is crucial to also keep track this step of the funnel.
That's why we've added an option to
add custom code snippets into your chatbot widget
. And now, you can:
  • Inject a
    Google Tag Manager
    script
  • Add a
    website analytics
    (eg. Plausible) code snippet
  • Programmatically
    interact with the chat
    widget elements
You can learn more about this feature in this guide. Check out our examples for common use cases.
Happy Thursday! Rafal from Chatwith here. I thought I'd let you know about exciting product updates we just shipped:
Claude-3
Claude Sonnet 3.5
New model is here! Since its release, Claude 3.5 Sonnet from Anthropic has been making waves. It's as capable (and more in some aspects) as GPT-4o from OpenAI. In our tests it turned out to be
faster and friendlier than GPT-4o - at the same cost
!
Starting today you can switch to this model if you're on one of our premium plans.
We recommend testing your chatbot after the change to make sure it works as expected on the new model.
Screenshot-2025-01-31-234840
Gemini 2.0 Flash
Released just a few days ago, this model instantly became a competitor to GPT-4o-mini from OpenAI. It is the fastest model in our offer, so if you're looking to wow your customers with
fast & smart responses
, I recommend giving this model a try!
Starting today you can switch to this model if you're on one of our premium plans.
We recommend testing your chatbot after the change to make sure it works as expected on the new model.
OpenRouter
From now on you can optionally use your own OpenRouter API key (instead of OpenAI) to bill your chatbot messages to. This helps with cost management, automatic fallbacks during downtime and paves the way to additional more alternative models in the future.
Small changes and fixes
Here's a list of other things that work better now:
  • New: Picing plans & add-ons
  • New: Add fallback utterance option ("default answer")
  • Improve: Increase chat footer text length
  • Improve: Upgrade Cohere Rerank to 3.5
  • Improve: Communicate adding members better
  • Fix: "undefined" instead of fallback utterance
  • Fix: Intermittent crash when loading conversations screen
  • Fix: Markdown list styling with dates
  • Fix: Hide autofocus warnings in browser console
  • Fix: Pages in errors not shown correctly in new sources modal
  • Fix: Handling of long conversations not fitting LLM context window
  • Fix: Issues when switching chatbots in dashboard
  • Fix: Doubled AI Action bubbles in chat
  • Fix: Saving email notifications schedule
Happy Thursday! Rafal from Chatwith here. I thought I'd let you know about some exciting updates we've just rolled out.
Share Conversations
"Go to archived, scroll to the third page... find the conversation with Jovial Penguin..."
- no more! From now on you can easily
share a direct link to a conversation with others
. Simply click the "Share" icon in the conversation toolbar to get a shareable link.
New Lead Notifications
Up until now, you could view leads collected by your chatbot only by visiting the conversation log.
We've added an option to enable
instant email and Webhook notifications whenever a new lead gets submitted
. Additionally, lead information is also attached to the usual periodic updates about your chatbot performance. Learn more about notification settings here.
Small changes and fixes
Here's a list of other things that work better now:
  • Fix: Do not show usage widget in sidebar to for Guests
  • Fix: Do not allow top domains in custom domain field
  • Fix: Preview not working in white-label domain if allowed domains are set
  • Fix: HTTP -> HTTPS redirect in white-label auth
  • Fix: “Manage Subscription” button not working
  • Fix: Improve parsing of OpenAPI specs
  • Fix: Do not allow Swagger 2.0 specs
  • Change: Remove legacy Customer Portals feature
  • Improve: Validate chatbotId in public API
Happy Thursday! Rafal from Chatwith here. I thought I'd let you know about some exciting updates we've just rolled out.
Frame 1000002223
Edit Sources
We've just added out a very simple, but powerful feature: an option to view and edit the content of your sources uploaded for training. No more guesswork what is included in chatbot knowledge. From now on, you are able to see exactly what data was used for training; and more importantly, easily add edits!
To get started, head to Knowledge section of your dashboard and click a "✏️" icon on any source.
Happy Thursday! Rafal from Chatwith here. I thought I'd let you know about some exciting updates we've just rolled out.
Frame 1000002222
Usage Analytics
We've created a more detailed breakdown of message usage per day and per model. Additionally, you can also view it for past billing periods. This will be definitely helpful if you need a closer look at your monthly usage! This analytics page can be accessed from the new "quick glance" Usage card in our updated sidebar.
Frame 1000002221
Popup & Bubble Settings
Besides configuring the chat icon position, icon and bubble text, now you can also decide when the bubble should disappear and optionally open the chat window automatically upon the user visit on your website. These two options were greatly anticipated and will help you increase engagement with your chatbot.
Small changes and fixes
Here's a list of other things that work better now:
  • Improvement: Change payment provider to Paddle (for new customers)
  • Improvement: Chat widgets load now 200% faster
  • Improvement: Add form validation in Refinements dialog
  • Improvement: Add hint about max Refinement length
  • Improvement: Unarchive and undelete conversation if a new message was added to it
  • Improvement: New sidebar with sections
  • Fix: Failed Action calls missing from conversation log
  • Fix: Bugs when switching chatbots
  • Fix: Characters usage attributed to another project member
  • Fix: Usage not counted in the last few hours of each month
Happy Tuesday! Rafal from Chatwith here. I thought I'd let you know about our biggest update yet:
landing-value-prop-team
Team Management
Now you can invite your teammates and clients to collaborate on a chatbot. No more juggling access credentials! With 3 new roles: Owner, Editor and Client you can delegate responsibilities and share access. Learn more here or go ahead and send some invites now.
New Customer Portal
Thanks to the new role-based access, team members with the Client role have now access to not only the conversation log, but also chatbot analytics and Refinements. With this upgrade, the old Customer Portal will be sunset soon. Tip: your dashboard can be also accessed through a Custom Domain, if configured.
If you currently use the old Customer Portal link, please migrate your stakeholders to the new role-based dashboard as soon as possible.
Small changes and fixes
Here's a list of other things that work better now:
  • Fix: Chatbot not recognizing new Action
  • Fix: Issues with access with multiple expired subscriptions
  • Fix: Some URLs not getting crawled on Safari
  • Fix: Refinements getting chunked incorrectly, resulting in large refinements not working
  • Fix: a crash when crawling some URLs
  • Improvement: Remove
    !important
    from chatbot launcher button CSS
  • Improvement: Avoid duplicated API calls to increase performance
  • Improvement: Show error if magic login email could not be sent
Happy Thursday! Rafal from Chatwith here. I thought I'd let you know about exciting product changes we shipped this week:
CleanShot 2024-08-14 at 2 
Messaging limits per Chatbot
This feature is a must have if you build custom chatbots for your clients! From now you can set a
maximum number of messages a chatbot can send in a month
. When a limit is reached, the chatbot becomes inactive. Learn more about this feature here.
Small changes and fixes
Here's a list of other things that work better now:
  • Improve: 2x faster chat widget loading
  • Improve: customize placeholders & button in lead form
  • Improve: background color of the lead form
  • Improve: Option to hide Action indicators in chat
  • Fix: OpenAI key validation
  • Fix: Intermittent "Something went wrong" error in chat
Load More