🍜 Changelog

​

5/15/23

• Integrations

  • Import documents directly from Google Drive or S3
    • Connect to Drive through OAuth flow, or paste in your service account keys to access your S3 files

• Azure OpenAI, Claude 100K model and Google PaLM models now available

• Misc

  • Additional settings for context variables like time decay
  • docx files now supported in uploads
  • Fixed issue with uploading csvs through UI would create duplicate rows
  • support for user identifier in completions API

​

4/24/23

• Document Search

  • Two new endpoints for searching documents
    • POST /api/datasets/{id}/search-documents - search documents with traditional keyword based search
    • POST /api/datasets/{id}/document-filtered-search - equivalent to search documents followed by a /search using the relevant documents as a metadata filter

• Slack & Discord deployments

  • Deploy an endpoint directly to a Slack workspace or Discord server in the deployments page!
  • Due to the nature of Slack & Discord apps, only one endpoint can be deployed at a time per workspace/server

• Misc

  • SSO support for Pro plan
  • Endpoint caching should improve endpoint performance
  • Security (PII protection) now available on all plans!
  • All filter types supported in App builder
  • Multiple URL upload

​

4/10/23

• Logging

  • Better observability into your API calls. See number of error and OK requests.
    

• Datasets

  • CSV file support
    To import/upload CSV files, ensure the correct columns already exist in the dataset. If some columns in the csv are meant for metadata, ensure they exist as metadata columns.

  • Auto embed image columns
    • Enabling auto embed on an image column will automatically generate a caption for the image and embed it.
      

• Security (Team plan)

  • Handle PII with confidence - we’ve added support for redacting, masking, or replacing detected PII.
  • Some included info types are SSNs, ICD9 codes, and credit card numbers. Let us know if theres one you’d like to see added.

​

3/28/23

• Self service sign up

  • No more waitlist! Log in with Google, Github, or email on the website
  • 7 day trial for all new users, then 35/month per seat

• Hybrid search (Team plan)

  • New hybrid search powered by instructor-large and SPLADE embeddings
  • Configure alpha in your app to weight semantics vs keywords during search.

• Misc

  • Several UX improvements (visualize forks)
  • Bug fixes (new API for uploading large files)
  • New URL! app.baseplate.ai

​

3/13/23 - Backend-as-a-Service for LLM Apps

• Hybrid Datasets 📑

  • Add any number of columns ( embeddings, data, metadata)
  • Use your datasets to provide your LLM with accurate and relevant context
  • Use the data returned via the API to create rich UX’s in your chat interface
  • You don’t have to manage a vector database AND a regular database. It’s all under one roof

• Document Management 📖

  • We’ve added a ton of support for managing datasets with multiple documents
  • It’s now extremely easy to replace vectors from an old document with content from new documents
  • You can now organize and return content based on the document id
  • Check out the API reference for some more info

​

3/2/23 - Chat GPT API 👀

​

OpenAI released the Chat GPT API, and we’ve added it Baseplate

• Go to the chat bubble from the top right of the playground to use our Chat GPT playground

• We’ve built in optimizations based on best practices to make your results better

​

2/24/23 - Introducing Baseplate for Teams

• Your Team, One Workspace

  • Your team’s prompt Templates, Datasets, and Playground+ sessions are now shared!
  • You can now work together on tasks like prompt iteration, configuring semantic search, and more.
  • Never lose track of your teams historical data. Look back and see which variants are performing best.

• Baseplate Endpoints

  • Found a variant that works? Deploy to production with a single API call
  • A variant consists of a prompt, a data augmentation (semantic search) resource, and a LLM
    * By composing different combinations, you can quickly experiment and find the best strategy for your use case
  • Configure your connection to a Baseplate endpoint once in code, then never worry again. You can run experiments and deploy new variants to that endpoint from the console.

​

2/20/23 - Baseplate 0.2

• Better Organization and Space Utilization. New Landing Page

  • Some of you asked for us to make text boxes more readable and simplify our navigation
  • There is now one sidebar (which you can minimize), so you have more space to work
  • It’s now much easier to read and compare prompts/completions
  • You also let us know that our web page didn’t express the value of the product. Check out our latest version!

• More Customized Embedding Support

  • You can now choose a chunk size
  • You can now choose a “k” factor. This changes the number of search results
  • You can now choose any number of variables to use in your embedding search

• Github Integration Alpha

  • You can now add our Github app to your repo and keep your templates in code and Baseplate in sync
  • This integrates with popular prompt formats, like Langchain’s serialization

• Pricing Update

  • The Starter tier is now $35 per seat
  • Teams looking to fine tune directly on the platform will need to upgrade to the Pro tier

• Bug Fixes

  • Way too many