Mastra•2mo ago

How to stop agent midway during stream while preserving the output streamed till that point

Hi Folks,

I was wondering if its possible to stop agent stream midway while also preserving the tokens that were streamed till the time user stopped the stream.

So essentially:

User sends a message.
Lets say agent streams about 200 tokens and user clicks on stop.
Agent stops generating and we save these 200 tokens to the memory.

Currently, if I stop the stream from frontend and abort using abort controller on the backend, it does not persist the message to the backend.

Mastra Triager•11/14/25, 11:35 AM

Created GitHub issue: https://github.com/mastra-ai/mastra/issues/10099

GitHub

How to stop agent midway during stream while preserving the output ...

This issue was created from Discord post 1438835729982689371: Hi Folks, I was wondering if its possible to stop agent stream midway while also preserving the tokens that were streamed till the time...

_roamin_•11/14/25, 11:33 PM

Hi @Kartik ! You should have a look at the option https://mastra.ai/reference/streaming/agents/stream#savePerStep

Reference: Agent.stream() | Streaming | Mastra Docs

Documentation for the method in Mastra agents, which enables real-time streaming of responses with enhanced capabilities.

KartikOP•11/15/25, 9:48 AM

Hey @Romain I checked savePerStep but afaik each "step" is a single LLM call.
My agents only make 1 step and if I stop the streaming midway within a step that step is lost.
I want partial steps to also persist to the database, similar to how chatgpt saves the tokens at the exact point at where you ask it to stop