2024-08-08: GPT-4o-mini support

This release includes support for gpt-4o-mini for the chat completion requests. This support required an update to the token counting package. See instructions for switching here:
https://github.com/Azure-Samples/azure-search-openai-demo/blob/main/docs/deploy_features.md#using-gpt-4
Unfortunately, gpt-4o-mini is only in eastus, so you may need to change the location of your OpenAI resource. You can change that in .azure/ENV-NAME/config.json

What's Changed

• Upgrade openai, openai-messages-token-helper, for gpt-4o-mini support by @dependabot in #1893

Full Changelog: 2024-08-07...2024-08-08

2024-08-07: Improved logging in prod, auth package upgrade

The most significant changes of this release are:

• Improved logging in production: you should now be able to see INFO and DEBUG level logs in production, if desired, as well as the HTTP access logs.
• Migration away from the python jose package for Entra token validation, as it was no longer maintained. Only relevant if you are using the optional authentication feature.

What's Changed

• New screenshot for README file. by @zedhaque in #1853
• Fixed link to private access markdown page by @bvdboom in #1859
• Fix logging by @pamelafox in #1874
• Update README.md code block formatting by @lamtrinhdev in #1878
• Rename Azure OpenAI key environment variable by @pamelafox in #1880
• Fix broken paths by @john0isaac in #1883
• Update deploy_features.md by @YIN-Renlong in #1885
• Replace python-jose with pyjwt by @blutril in #1875

New Contributors

• @bvdboom made their first contribution in #1859
• @lamtrinhdev made their first contribution in #1878
• @YIN-Renlong made their first contribution in #1885
• @blutril made their first contribution in #1875

Full Changelog: 2024-07-22...2024-08-07

2024-07-22: Removal of VM from private network deployment

We previously included Bicep to enable the creation of a VM and Bastion for connection to a privately deployed app endpoint. However, due to security concerns around the VM enabling access to the endpoint, we've removed that option from deployment. To access the endpoint, you can instead use an Azure VPN gateway, Azure Virtual Desktop, other means of accessing the VNet, or add an allowance for your IP to the chat app's firewall.

What's Changed

• fix validation pipeline with missing files and actions by @anfibiacreativa in #1774
• Bump the python-requirements group across 1 directory with 21 updates by @dependabot in #1851
• Bump @fluentui/react-icons from 2.0.221 to 2.0.249 in /app/frontend by @dependabot in #1848
• Bump marked from 13.0.0 to 13.0.2 in /app/frontend by @dependabot in #1847
• Bump @azure/msal-react from 2.0.6 to 2.0.21 in /app/frontend by @dependabot in #1849
• Remove provisioning of VM and Bastion by @pamelafox in #1852

New Contributors

• @anfibiacreativa made their first contribution in #1774

Full Changelog: 2024-07-16...2024-07-22

2024-07-16: Seed parameter, Azure Verified Modules

This release includes:

• the addition of a seed parameter. That is particularly helpful if you are evaluating your app.
• a partial migration to Azure Verified Modules to reduce the amount of custom Bicep in this repo and inherit security best practices automatically. We will continue to migrate as many modules as possible.
• an improvement to the prepdocs script so that you can pass "--removeall" and "--remove" on the CLI. You still need to modify the script to specify the filename, however.
• a fix for the local frontend server, to address an issue with running "npm run dev" inside Dev Containers

What's Changed

• Bump typescript from 5.4.5 to 5.5.3 in /app/frontend by @dependabot in #1796
• Docs: More ideas for improving search results by @pamelafox in #1811
• Fix vite by adding host by @pamelafox in #1812
• Prepdocs: Support additional args by @mattgotteiner in #1813
• Use AVM for log analytics and application insights. by @tonybaloney in #1817
• Bump @vitejs/plugin-react from 4.1.1 to 4.3.1 in /app/frontend by @dependabot in #1757
• Add an optional seed parameter by @pamelafox in #1814
• Bump certifi from 2024.6.2 to 2024.7.4 in /app/backend by @dependabot in #1789
• Use AVM for all Cognitive Services accounts and services by @tonybaloney in #1818

Full Changelog: 2024-07-10...2024-07-16

2024-07-10: Responsive design for mobile

This release includes improved CSS and a burger menu for smaller viewport sizes, so that the app works better for mobile phones. Please test out and report any issues.

What's Changed

• CSS changes for responsive design by @zedhaque in #1646

New Contributors

• @zedhaque made their first contribution in #1646

Full Changelog: 2024-07-03...2024-07-10

2024-07-03: Authentication and security improvements

The most significant change in this release is improvements to the login system to handle token refresh in the browser for users who leave the tab open for long periods of time (PR #1778). Please bring those changes in if you're using auth.
 

What's Changed

• Remove powershell feature by @pamelafox in #1752
• Support use of AzureOpenAI proxy by prepdocs by @pamelafox in #1760
• Fix arg specification by @pamelafox in #1761
• Set ACLs to Deny for storage by @pamelafox in #1765
• fix ruff check by @john0isaac in #1772
• Bump the python-requirements group across 1 directory with 19 updates by @dependabot in #1775
• Fix login state management by @mattgotteiner in #1778
• Make OpenAI deployment optional by @pamelafox in #1777

Full Changelog: 2024-06-20...2024-07-03

2024-06-20: Settings allow semantic ranker with vector search

The most significant change in this release is that you can try out semantic ranker with vector search, in case that's better for you than semantic ranker with hybrid search. In our tests, semantic+hybrid is always the best, but we want to make sure you can test all the combinations.

Various packages and documentation pages have also been updated.

What's Changed

• Switch to PSRule for security testing by @tonybaloney in #1687
• UI developer settings fix by @vedenev in #1688
• Updates to Markdown files (links, lint, wording) by @pamelafox in #1700
• Allow semantic ranker with vector search by @pamelafox in #1701
• Bump azure-identity from 1.16.0 to 1.16.1 in /app/backend by @dependabot in #1712
• Bump the node-packages group across 1 directory with 20 updates by @dependabot in #1706
• Bump typescript from 5.2.2 to 5.4.5 in /app/frontend by @dependabot in #1717
• Bump react and @types/react in /app/frontend by @dependabot in #1718
• Bump marked from 9.1.6 to 13.0.0 in /app/frontend by @dependabot in #1719
• Bump react-router-dom from 6.18.0 to 6.23.1 in /app/frontend by @dependabot in #1721
• Bump @azure/msal-browser from 3.10.0 to 3.17.0 in /app/frontend by @dependabot in #1720
• Fix tooltips typo by @pamelafox in #1736
• Update deploy_private.md with video link by @pamelafox in #1738
• Bump urllib3 from 2.2.1 to 2.2.2 in /app/backend by @dependabot in #1739
• Bump ndjson-readablestream from 1.0.7 to 1.2.0 in /app/frontend by @dependabot in #1731
• Bump @types/react-syntax-highlighter from 15.5.9 to 15.5.13 in /app/frontend by @dependabot in #1732
• Add support for specifying an Azure OpenAI Key by @pamelafox in #1746
• Bump react-dom and @types/react-dom in /app/frontend by @dependabot in #1734

New Contributors

• @vedenev made their first contribution in #1688

Full Changelog: 2024-06-05...2024-06-20

2024-06-05: Conform app to new AI Chat Protocol

The big change in the latest release is that the app now confirms to the official Microsoft AI Chat Protocol, documented here:
https://github.com/microsoft/ai-chat-protocol/tree/main/spec#readme
There is also a JS SDK for that protocol that we will adopt soon to simplify our frontend parsing code.
You can see the pull request for more details, but basically:

• Instead of sending "stream": True to get a streaming response, we send to a different path, "/chat/stream".
• Instead of passing down the full choices list from the OpenAI response, we pass down only the first choice, specifically it's message or delta.

We have also updated the evaluator tool to work with either the new version of the backend protocol or the old, configurable via JMESPath expressions in the config JSON.

This change also includes a revamp of the Developer Settings to include tooltips, to try and make it more clear what each setting does.

What's Changed

• Bump the python-requirements group with 26 updates by @dependabot in #1673
• Update deploy_features.md by @hadirgax in #1689
• Add clickable help icons for developer settings by @pamelafox in #1522
• Upgrade to latest version of AI Chat Protocol by @pamelafox in #1682

New Contributors

• @hadirgax made their first contribution in #1689

Full Changelog: 2024-06-03b...2024-06-05

2024-06-03b: Speech output with browser

You can now optionally enable speech output using the Browser SDKs, as opposed to using the Azure Speech SDK. The speech output is generally not as fluid, but it's entirely free, so it's a great zero-cost option to increase accessibility. See docs at:
https://github.com/Azure-Samples/azure-search-openai-demo/blob/main/docs/deploy_features.md#speech-output

We've also added documentation about a Python-based load balancer for OpenAI instances. See:
https://github.com/Azure-Samples/azure-search-openai-demo/blob/main/docs/productionizing.md#openai-capacity

 ## What's Changed

• Update productionizing.md by @simonkurtz-MSFT in #1677
• Enable diagnostics for search service by @pamelafox in #1676
• feat: add low cost browser text to speech output using WebSpeechAPI by @john0isaac in #1671

New Contributors

• @simonkurtz-MSFT made their first contribution in #1677

Full Changelog: 2024-06-03...2024-06-03b

2024-06-03: GPT-4O support for optional vision feature

The optional vision feature now uses GPT-4O instead of GPT4-turbo-with-vision, due to much better performance.

Unfortunately, due to the difference in region availability, it may be difficult to switch to GPT-4O in your existing resource groups, depending on what region you originally picked. You may need to change the region for the OpenAI resource group in .azure/ENV-NAME/config.json, delete the previous OpenAI instance, and deploy a whole new OpenAI instance.

We did discover one vision-related splitting issue while debugging this change.

See pull request for performance statistics and more details on the change.

What's Changed

• Fix overlap percentage calculation by @bastbu in #1668
• Bump the python-requirements group across 1 directory with 17 updates by @dependabot in #1665
• Use gpt-4o for vision approach by @pamelafox in #1656

New Contributors

• @bastbu made their first contribution in #1668

Full Changelog: 2024-05-29...2024-06-03