Deepseek R1 Agent

Guys you gotta keep going fast!!! Cline added Deepseek R1 agentic mode one day after it was out. You guys did not even add deepseek v3 yet after weeks that it has been out. What’s going on?!?!?

Agentic mode Deepseek R1 is super needed. LETSGOOOO

44 Likes

thats what im saying

5 Likes

requested here aswell

3 Likes

cast the first vote, let’s goo!!

3 Likes

They did in fact add deepseek-v3, and they just added deepseek-r1. Have you checked your settings, under models in Cursor?

2 Likes

Unfortunately, DeepSeek can not currently be used with agent mode, which is the request here, but we are investigating if we can add it soon!

15 Likes

as a no-code, i really need that future! please please please

I support the request, it would be really good.

please add it as fast as you can

yeah, we need it in the agentic mode. even ready to use after taking the responsiblity of ‘non-safe’ characteristics of the deepseek models to myself

2 Likes

Would be awesome it least make R1 and Sonnet work together, where R1 is responsible for reasoning and planning, and Sonnet for execution.

3 Likes

I have a question:

If Cursor has OpenAI and Anthropic as partners, would that influence their model choices? (maybe they might boycott “non-US” models?)
(just thinking aloud)

4 Likes

R1 is an open-source model that can be locally deployed, with electricity costs constituting the majority of its operational expenses. I believe Cursor will make choices that benefit both users and itself – this is a double win scenario.

1 Like

Need to test two cursor agents opening the same context vscdb and assigning each to a diff model, but shared context - and have them work in step by reading and writing to a steps.rmd whereby they perform task, update steps.rmd with a readme chunk with development_diary format, and timestamps, iteration UUID and a backup_checkpoint saved out - then the next model takes the next step with the steps.rmd as each agentic action’s starting input context.

3 Likes

I’m wondering about the same… maybe it is a partnership issue there, althought they already included in the chat, not sure why they can’t use it in the agent mode as well if that is less expensive, maybe they need to deal that part with Deepseek

THIS IS A MUST! why do you only allow a couple LLM’s to be used with composer. I would REALLY like to be able to use gemini in composer with my own api after i’ve used up the 500 “fast” requests, which usually happens within the first week of the month every month… and your unlimited slow requests is a scam, so i was hoping to use gemini for free, but in agentic mode.

gemini supports function calling, if you werent aware.

3 Likes

Just to update everyone, the team is working on this and will hopefully make it available soon!

27 Likes

In the meantime, would it work to just give Claude a tool to send a prompt to deepseek and read the response?

1 Like

Awesome, thanks you for this feedback @danperks , and thanks team Cursor :smiling_face_with_three_hearts:

but the question is, since deepseek is not supporting images yet will we still able to use compose feature with image? also deepseek pricing is really cheap compare to sonet, would there be any changes to the pricing aswell?