[Bug] @microsoft/teams-ai v1.7.4 doesn't work anymore with streaming responses.

[sosafe-jose-gonzalez]

Hola :)

I got a bot running with @microsoft/teams-ai v1.7.4

Everything has been running fine for the past few months but since Friday we have errors when doing streaming responses.

RestError: Only start streaming and continue streaming types are allowed as a typing activity

This looks like there was a change on the API?

An example of the code:

const streamingResponse = new StreamingResponse(context);
streamingResponse.setGeneratedByAILabel(true);

streamingResponse.queueInformativeUpdate(text);

// Calls to an AI that will stream a response
// inside a for loop
streamingResponse.queueTextChunk(text)

await streamingResponse.endStream();

// finishes loop

The solution for now was to disable streaming which is not optimal for end users.

Is there something that can be done while running v1 of the SDK or is the solution is to migrate to v2?

[lilyydu]

Hi @sosafe-jose-gonzalez, which version are you running with now? 1.7.4 is still the latest version for TS in our v1 SDK

[sosafe-jose-gonzalez]

Hi @lilyydu I'm running 1.7.4 on the v1 SDK.

Basically whenever calling streamingResponse.setGeneratedByAILabel(true);, streamingResponse.queueTextChunk(text);, or await streamingResponse.endStream(); I get RestError: Only start streaming and continue streaming types are allowed as a typing activity

[lilyydu]

Hi @sosafe-jose-gonzalez,

Are you queueing an informative update before sending your text chunks? The stream activities all need to be referencing the same activity ID which comes back after the first update is sent

Could you also double check you're following the rules here:
https://learn.microsoft.com/en-us/microsoftteams/platform/bots/streaming-ux?tabs=csharp#stream-messages-user-experience

[sosafe-jose-gonzalez]

Hi @lilyydu as my example above, this is what I was doing up to 2 weeks ago when it was working and then it just stopped working.

When a create a new StreamingResponse with the context I get a stream ID, which is the activity ID.

The only thing maybe I'm not sure is when to set the AI label, I'm doing it after creating a streaming response but that was not an issue before.

const streamingResponse = new StreamingResponse(context);
streamingResponse.setGeneratedByAILabel(true);

streamingResponse.queueInformativeUpdate(text);

// Calls to an AI that will stream a response
// inside a for loop
streamingResponse.queueTextChunk(text)

await streamingResponse.endStream();

// finishes loop

[lilyydu]

@sosafe-jose-gonzalez Let me check with our Bot API team, and get back to you. It should be fine to set the AI label before

[lilyydu]

@sosafe-jose-gonzalez Could you provide this information and we can trace our service logs?

• MS-CV/conversation + timestamp of failure
• botId + tenantId + timestamp of failure

[sosafe-jose-gonzalez]

conversationId - a:1TWnnJDiaf_f8qmrRZsdzkHVfKmC8P8eLBq5-4C0x64tjCISpE0up1TKWz9mjbEHUy-GpmYOYVZ3bn4u3BMiZBfaDtGKXh5yqcD80amfL6yls-aRN0XDMLkD403BeP89J

botId - eca88a84-5055-46eb-9bbb-1c08d988434a

tenantId - 558c21fb-a59b-4b04-abd3-a7c0c8eccc8a

timestamp - 1759420705181

[lilyydu]

Hi @sosafe-jose-gonzalez,

They look into the logs and it seems that the very first POST request was accepted, but was not recognized as a streaming activity. A few ms later, another POST to the same activity had the streaminfo entiy but the validation failed. We will need to check the payload that is being sent to the bot - can you provide us what your activities sent look like?

I also noticed you're sending the same text as the informative update and text chunk - you'll want to send your actual text result as a text chunk, the informative is what shows up when it's loading

[sosafe-jose-gonzalez]

I just send this, what payload would you need? the context?

const streamingResponse = new StreamingResponse(context);
streamingResponse.setGeneratedByAILabel(true);

streamingResponse.queueInformativeUpdate('informative text');

// Calls to an AI that will stream a response
// inside for loop
streamingResponse.queueTextChunk('streamed response from AI')
// end for loop

await streamingResponse.endStream();

[corinagum]

@sosafe-jose-gonzalez Please send the activity payload of the constructed stream.

[heyitsaamir]

This regression is due to botbuilderjs migrating to azure/core-client. Essentially, for an activity that looks like this:

{
   "type":"typing",
   "text":"foo",
   "channelData":{
      "streamType":"streaming",
      "streamSequence":1
   },
   "entities":[
      {
         "type":"streaminfo",
         "streamType":"streaming",
         "streamSequence":1
      }
   ]
}

The resultant payload is this:

{
   "type":"typing",
   ...
   "text":"foo",
   "entities":[
      {
         "type":"streaminfo"
      }
   ],
   "channelData":{
      "streamType":"streaming",
      "streamSequence":1
   },
   "replyToId":"1761610592210"
}

Notice how "streamType" got dropped in entities?

I believe this is fixed in microsoft/botbuilder-js#4903, so once that's released, we'll pick it up and that should fix this issue. Alternative could be to revert to an older version of the library that is not dependent on the change to azure/core-client