Stephen (Dust)

hi Piyush Nimoria looks like you're hitting a timeout of 30secs on Axios. If the call you're making to Dust is a "Talk to an agent" you could try to make sure that the agent is faster by using a faster model like GPT 4o-mini for example

Hi Gilles, that's definitely doable, and one of the common use cases of Dust's API 🙂 This is usually achieved by using the conversation endpoint directly, or the javascript SDK In terms of timeout, Dust will cut at 30s in synchronous mode. (setting is called blocking : true) If your agents are taking longer than that to reply, try the asynchronous mode where you can poll dust to get your events. Hope this helps!

💡 nice one! It's not baked in the system right now, but with the recent issues with Anthropic, we're actively trying to find a smooth way to fallback to a working model when these situations happen 🙂

If the labels are not synced from the Dust native integration, I think you'll have to edit the files after they've been added to Dust, adding the tag programatically, using the upsert document endpoint

Hi Denis Kirchen 👋 I think you could use the filter feature and add the account name as a new label in this use case! Hope this helps! https://www.youtube.com/watch?v=QwHyMpVJva4&ab_channel=Dust-Transformhowworkgetsdone.

Léandre Le Bizec I think you've been using n8n as well, right? Is this a problem you are also having ?

Hi Hélène Bonnet 👋 Correct David Ebbo,thanks for chiming in ! Indeed, if the user already has a profile on dust.tt, they cannot join a workspace on eu.dust.tt with the same identity. The workaround would be to use another email address

Hi Yohann Bosco! thanks for reporting this. This means that Confluence was rate-limiting your Dust workspace on this specific Confluence space. I'm not sure what the current maximum rate is on Confluence, but retrying later should solve it (from the log of your action, retrying after 1minute was enough). I'm expecting that only large operations (like indexing for the first time) could result in rate limits, and once the first batch is done, the connected source should operate normally. Let us know how that goes after a retry !

Hi Romain Dechamps! I ran a few tries on a document retrieval agent today with Claude Sonnet 3.7 with the Creativity setting set to factual, and gave it instructions based from the template @techDocDigger With this setup, the agent seems to consistently be helpful and reference relevant sources from the datasource. Happy to take a look at your config with you if this does not make your agent work, DM me if I can help !