Improving Response Times for Content Redaction Agents

·

Hey team! I was wondering if there was a way to reduce the time between an agent's call and the response. Sometimes the delay is more than 2-3 minutes for an content redaction agent using Claude 3.7 or Claude4. Is there a way to reduce this dust delay? Improve in the prompts, setups or using another LLM?

5 comments

· Sorted by newest

Remi
·
·
Etienne Dubois indeed, lags are often caused by delay in Claude's servers. Switching to the more recent gpt5 model should help. It's fast and smart ! Making sure the agent doesn't call tools when it doesn't need to can help too. You can do that through the instructions (eg: "never use tools unless asked")
❤️1
Etienne Dubois
·
·
Thanks Hugo! Indeed, we try to divide tasks between several agents to avoid this 🙂
Hugo TAVERNE
·
·
Hi Etienne, it might be coming from the database your agent is pulling from. If you multiply the amount of content (website, Notion docs, etc.), the agent might take longer to process everything for each new task.
❤️1
Etienne Dubois
·
·
Thanks Melvin for your answer! So mainly switching the LLM? Nothing on your side regarding setups or prompt improvements?
Melvin Duveau
·
·
When I need a fast answer, I am calling GPT 4.

Loading comments

Remi
·
·
Etienne Dubois indeed, lags are often caused by delay in Claude's servers. Switching to the more recent gpt5 model should help. It's fast and smart ! Making sure the agent doesn't call tools when it doesn't need to can help too. You can do that through the instructions (eg: "never use tools unless asked")
❤️1
Etienne Dubois
·
·
Thanks Hugo! Indeed, we try to divide tasks between several agents to avoid this 🙂
Hugo TAVERNE
·
·
Hi Etienne, it might be coming from the database your agent is pulling from. If you multiply the amount of content (website, Notion docs, etc.), the agent might take longer to process everything for each new task.
❤️1
Etienne Dubois
·
·
Thanks Melvin for your answer! So mainly switching the LLM? Nothing on your side regarding setups or prompt improvements?
Melvin Duveau
·
·
When I need a fast answer, I am calling GPT 4.