๐ค Fixed bugs and added support for Deepseek-R1 model
๐ค Added OpenRouter Deepseek-R1 model
๐ Fixed bug: OpenAI compatible providers (including OpenAI) didnโt return a proper error response message - only โWhoops! We couldnโt send your message. Let us know if this keeps happening.โ
๐ Fixed bug: Test inputs (rows) could not be deleted if some columns were hidden
๐ค New Model Support, Magic buttons, Testing Improvements, and Onboarding
๐ค Added support for the Claude 3.5 Haiku model from Anthropic.
โจ Introduced several โmagicโ buttons throughout the app to improve your experience with LLMs. These include prompts for enhancing prompts, generating test cases, and more! Let us know which ones you find and how useful they are - itโs like an Easter egg hunt!
๐ Fixed an issue where the test data grid couldnโt delete repeated test cases.
โ๏ธ Introduced a โnoneโ option for tool choice.
๐ Added a history viewer for better visibility into past events.
๐ฑ Improved onboarding experience for mobile devices.
๐ At the core of Langtail 1.0 is our new spreadsheet-like interface for testing LLM applications, designed to feel natural for anyone used to working with Excel or Google Sheets.
๐ป Your prompts can now call tools and have the code run directly within Langtail, making prototyping and testing much easier without the need for mocking responses.
๐ Enjoy secure and fast code execution in a sandboxed environment powered by QuickJS.
๐ง Langtail 1.0 introduces Assistants โ stateful entities that automatically manage memory and conversation history, reducing the need for additional code.
๐ Assistants can be used across models, in tests, deployed as APIs, or integrated with tools.
๐ ๏ธ Tool Calls in Assistant Playground and Gemini Support
๐ ๏ธ Added the ability to handle tool calls in the Assistant Playground, allowing you to provide responses for tool calls not handled by Langtail function handlers.
๐ค Introduced native support for Gemini models, leveraging the @google/generative-ai library under the hood, with support for tools, streaming, and various media formats as input.
๐ Expanded Message Types and Provider Visibility
๐จ Added support for other message types, tools, and images in the โmessagesโ column, allowing you to simulate user and assistant messages, as well as mock tools and upload images for Vision testing.
๐ Unlocked providers are now displayed at the top for better visibility.
๐ค Introducing Anthropicโs Claude 3.5 Sonnet Model
๐ Added support for Anthropicโs Claude 3.5 Sonnet model, offering faster and cheaper performance compared to Claude 3 Opus, with enhanced reasoning, knowledge, and coding abilities.
You can now add additional columns that act as helper variables for assertions (JavaScript, LLM assertions). Place the correct answer in these helper variables and use them in assertions for more efficient and accurate testing.
Based on user feedback, weโve improved LLM assertions. Now, the entire conversation history is included in the context, providing a more comprehensive view for better testing.
๐จ Multiple Messages in the โMessagesโ Column ๐จ
You can now use multiple messages in the โmessagesโ column. Currently, itโs possible to simulate user and assistant messages. Weโre working on the ability to mock Tools and upload images for Vision testing.Tests are a big focus for us, and weโre committed to continuously improving them.
Stay tuned for more updates!
๐ Upcoming Feature: LLM as a Judge Assertion ๐
Stay tuned! Weโre in the process of developing an โLLM as a judge assertionโ feature, set to take our testing capabilities to the next level. More updates coming your way soon!
Weโre thrilled to introduce our new dashboard view! Now serving as the main hub for each project, the dashboard provides a snapshot of your projectโs activity and performance at a glance. With recent logs and quick statistics, you can stay updated the moment you log in.
The first version of the Langtail SDK is now live! ๐ฅณ The SDK can function as an OpenAI proxy and is also compatible with prompts directly deployed within Langtail. Dive into the details over at our GitHub docs.Currently, weโve rolled out the JS SDK. If youโre interested in a Python SDK, weโd love to hear from you.
Curious about the team behind Langtail? Visit our newly updated About Us page to learn more about our mission and the people who make Langtail possible. Learn more
Weโve been busy expanding our documentation to help you get the most out of Langtail. Head over to our docs page to explore the latest information and guides.
Already building your application using the OpenAI SDK? No problem! With Langtail Proxy, you can experiment with Langtail without migrating your prompts. Hereโs how:
Create your Langtail account
Generate your project API key
Modify your code by updating the base URL to "https://proxy.langtail.com/v1 and passing the Langtail API key
Chatbots can now stay up-to-date with the current date! Introducing the $date helper, a new feature designed to keep your chatbots in sync with real-time.
Ever wondered about our pricing? Weโve just launched our pricing page, making it easier for you to find the perfect plan for your needs. Check it out!
Good news, everyone! Weโve just launched a new feature that lets you publicly share your prompts. Previously, users needed an account to view a shared prompt, but now, anyone can view it thanks to our new read-only Playground view. ๐
In the read-only Playground, you can see all messages, variables, and Tools (functions). Itโs an open window to the world of your prompts. Currently, only a fixed version of the prompt can be shared, but weโre already working on updates that will let you share and always view the most recent version of a prompt.
Weโre excited to announce that weโve released support for a templating engine, based on Handlebars! With this new feature, you can now tackle more complex use cases, including:
Conditional parts of prompts
Variations of prompts
Simply deploy a prompt with this template, send the correct variable, and watch as the final version of the prompt resolves. You can then view this in the Logs. This added flexibility opens up a world of possibilities for your prompt handling.
Managing your projects just got easier! Weโve added a new item in the project-level sidebar that lets you perform common actions like renaming or deleting a project.
Your workspace is now even more flexible! Weโve refined the settings to allow you to rename your workspace and made the UI more consistent across the different sections for a smoother experience.
Weโve completely redesigned our Logs for a cleaner, more user-friendly interface. Plus, Logs now also display the JSON representation of whatโs being sent from Langtail to end providers like OpenAI. We think youโll love this upgrade as much as we do!
Weโre thrilled to announce that support for the new OpenAI Tools is here! Functions now visually resemble TypeScript functions and are even typesafe. The types are generated based on the JSON schema. Weโre confident youโll find them incredibly useful.
Weโre thrilled to announce that GPT-4 Vision is now live in production! You can now input images along with text, opening up a whole new realm of creative use cases. For instance, take a photo of an invoice and ask for the cost of each item โ Vision has you covered.
Want to give Vision a spin? Head over to our Playground and upload an image or insert an image URL. Please note that Vision takes a few seconds to process, so a bit of patience is required.
Weโve also rolled out a couple of key updates for improved user experience:
New Workspace Settings: Now, you can manage your provider API keys and team members all in one place.
Refined UX for Per API Budget: Based on your feedback, weโve refined the UX and terminology for the per API budget. You must now enable the budget before setting the total for your API key and specific user allowances.