Need Help with Script for Dust SSE Conversation Management?

·Jun 04, 2025 10:25 AM·

Hello everyone, does anyone have a standard script to create a conversation, send a message, listen for agent response and retrieve it ? I’ve been trying to create such a script without success and having difficulty understanding how Dust SSE works.

5 comments

· Sorted by Oldest

Jean Langlois-Meurinne
·
·
Loading attachments
Jean Langlois-Meurinne
·
·
Voici le fichier sur lequel j’ai travaillé pour l’instant mais tout modèle qui réalise ces fonctions de base serait utile pour que je comprenne bien le fonctionnement des flux SSE
David Ebbo
·
·
Hi Jean. Check out this code, which does something like this. It's not standalone code, as it's a plugin to the LLM CLI (https://github.com/simonw/llm) but the logic to stream events should not be much different. https://github.com/davidebbo/llm-dust/blob/main/llm_dust.py
Jean Langlois-Meurinne
·
·
Thanks, it works even if sometimes there is a timeout before any complete answer is received (even with very long response time). I’ll try to work it out and tell you if it doesn’t
David Ebbo
·
·
Does your agent by any chance run tools that need to be approved? If so, try passing "skipToolsValidation": true when you create the conversation.

Jean Langlois-Meurinne
·
·
Loading attachments
Jean Langlois-Meurinne
·
·
Voici le fichier sur lequel j’ai travaillé pour l’instant mais tout modèle qui réalise ces fonctions de base serait utile pour que je comprenne bien le fonctionnement des flux SSE
David Ebbo
·
·
Hi Jean. Check out this code, which does something like this. It's not standalone code, as it's a plugin to the LLM CLI (https://github.com/simonw/llm) but the logic to stream events should not be much different. https://github.com/davidebbo/llm-dust/blob/main/llm_dust.py
Jean Langlois-Meurinne
·
·
Thanks, it works even if sometimes there is a timeout before any complete answer is received (even with very long response time). I’ll try to work it out and tell you if it doesn’t
David Ebbo
·
·
Does your agent by any chance run tools that need to be approved? If so, try passing "skipToolsValidation": true when you create the conversation.