Mastra•5mo ago

Reasoning part not saved in Mastra Memory message history?

Is reasoning not being saved? I get reasoning streamed in during generation, but when inspecting hte mastra messages table, I only see the text and tool invocation parts.

I'd expect any data part to just be saved to the DB, as long as it's not

transient: true

transient: true

transient: true

transient: true.

EDIT: okay I see the issue. For some reason, it gets saved like this..

{
  "format": 2,
  "parts": [
    {
      "type": "reasoning",
      "reasoning": "",
      "details": [
        {
          "type": "text",
          "text": "The user is asking me to ..."
        }
      ]
    }
  ]
}

{
  "format": 2,
  "parts": [
    {
      "type": "reasoning",
      "reasoning": "",
      "details": [
        {
          "type": "text",
          "text": "The user is asking me to ..."
        }
      ]
    }
  ]
}

{
  "format": 2,
  "parts": [
    {
      "type": "reasoning",
      "reasoning": "",
      "details": [
        {
          "type": "text",
          "text": "The user is asking me to ..."
        }
      ]
    }
  ]
}

{
  "format": 2,
  "parts": [
    {
      "type": "reasoning",
      "reasoning": "",
      "details": [
        {
          "type": "text",
          "text": "The user is asking me to ..."
        }
      ]
    }
  ]
}

The

reasoning

reasoning

reasoning

reasoning property seems to be an empty string.

I'm on AI SDK v5 beta and the latest mastra beta release.

Is there a utility function that converts the Mastra history to a Vercel AI SDK v5 UI message array?

I'm not calling Mastra via

useChat

useChat

useChat

useChat, but instead I use the memory API to fetch the message history.

Ariel WeinbergerOP•7/28/25, 9:48 PM

I think I found it -

    const { uiMessages } = await memory.query({
      threadId: artifactId,
      resourceId: projectId,
    });

    const { uiMessages } = await memory.query({
      threadId: artifactId,
      resourceId: projectId,
    });

Ariel WeinbergerOP•7/28/25, 9:51 PM

Reasoining data parts are still incompatible with v5.

They're missing the statestate property ("done" for example).

Ariel WeinbergerOP•7/28/25, 10:02 PM

can I somehow intercept the messages before they get saved to memory? I need to add my own metadata to certain messages.

Tyler•8/1/25, 6:33 PM

@Ariel thanks for the report! I think this might be an issue when we're converting from "format: 3" (which should be saved in the db) to "format: 2" in the json you pasted.
For intercepting/modifying in a middleware-like API, we're planning on adding that very soon, working on it next week

Tyler•8/1/25, 6:35 PM

can you open a GH issue with some steps on how to reproduce it?

Mastra Triager•8/5/25, 12:17 AM

Created GitHub issue: https://github.com/mastra-ai/mastra/issues/6532

GitHub

[DISCORD:1399493184358318274] Reasoning part not saved in Mastra Me...

This issue was created from Discord post: https://discord.com/channels/1309558646228779139/1399493184358318274 Is reasoning not being saved? I get reasoning streamed in during generation, but when ...

Abhi Aiyer•8/5/25, 5:18 PM

hi @Ariel we're working on our v5 support in a more backwards compatible way. We did find this exact issue in this work so will be fixed soon!

I posted an update at the top of the mega issue our plans for better v5 compat https://github.com/mastra-ai/mastra/issues/5470

GitHub

AI SDK v5 Support mega issue · Issue #5470 · mastra-ai/mastra

AI SDK v5 has been released!

Interim We've also released new @ai-v5 pre-release packages so you can try out v5 in Mastra!

The new pre-release packages contain a backwards compatibility laye...

Reasoning part not saved in Mastra Memory message history?

Similar Threads