DeepSeek(3rd.) - Your conversations, more than seven days. · A third-party Kotlin client for DeepSeek
·(edited)·· reads0
AI TranslationSimplified ChineseEnglish
Key Insights
DeepSeek open-sourced their official chat demo a few days ago, and almost immediately announced they'd be throttling conversations in the DeepSeek App because they were short on compute. Over the past couple of days they've tightened the restriction even further, making it hard to hold long conversations in the client. Honestly, I don't think the policy is a problem. It's been out for days and I haven't noticed any impact on my own usage — the DeepSeek App's search tool is excellent, far more accurate than the results most agent-search APIs give you, so I basically reach for the DeepSeek client whenever I need to look something up on my phone. And that kind of research-style usage rarely leads to long multi-turn conversations, so I never felt the squeeze. I also couldn't find anyone quickly turning the official demo into a third-party app — so I did it myself.
There are already quite a few projects out there — DSH GO, for instance, is fairly well known these days. But even though dsh go already exists, it leans on deepseek harness as its engine, and for folks like us who just want a plain daily driver, having to fire up a big chunky thing every time isn't exactly ideal. So I rolled up my sleeves.
To keep it close to the original while staying tiny and lightweight, I built it in Kotlin. Qwen4-Max just came out and was on sale, so I figured I'd give it a spin.
I've been pretty swamped lately and can't be bothered to walk through the whole process, so let me just share what I've gotten done over the past few days.
So far it's a fairly complete recreation of the official app — voice input and voice output are both done, and I've taken a first pass at conversation editing and conversation forking. It's only 3.7 MB right now, and I don't think future additions will push it past 5 MB. That's what I call small and beautiful. I've promised to keep all data stored locally, though, so more conversations and more images will inevitably take up some space — nothing I can do about that one.
DeepSeek open-sourced its official Chat Demo and limited long conversations in the app due to insufficient compute. The author thinks this has little impact and finds the search tool useful. Since existing projects like DSH GO rely on the deepseek harness and are too large for everyday use, the author built a third-party app in Kotlin. It now closely replicates the official app, supporting voice input and output, message editing and forking, at just 3.7MB. Data is stored locally, delivering a small yet polished experience.
DeepSeek open-sourced their official chat demo a few days ago, and almost immediately announced they'd be throttling conversations in the DeepSeek App because they were short on compute. Over the past couple of days they've tightened the restriction even further, making it hard to hold long conversations in the client. Honestly, I don't think the policy is a problem. It's been out for days and I haven't noticed any impact on my own usage — the DeepSeek App's search tool is excellent, far more accurate than the results most agent-search APIs give you, so I basically reach for the DeepSeek client whenever I need to look something up on my phone. And that kind of research-style usage rarely leads to long multi-turn conversations, so I never felt the squeeze. I also couldn't find anyone quickly turning the official demo into a third-party app — so I did it myself.
There are already quite a few projects out there — DSH GO, for instance, is fairly well known these days. But even though dsh go already exists, it leans on deepseek harness as its engine, and for folks like us who just want a plain daily driver, having to fire up a big chunky thing every time isn't exactly ideal. So I rolled up my sleeves.
To keep it close to the original while staying tiny and lightweight, I built it in Kotlin. Qwen4-Max just came out and was on sale, so I figured I'd give it a spin.
I've been pretty swamped lately and can't be bothered to walk through the whole process, so let me just share what I've gotten done over the past few days.
So far it's a fairly complete recreation of the official app — voice input and voice output are both done, and I've taken a first pass at conversation editing and conversation forking. It's only 3.7 MB right now, and I don't think future additions will push it past 5 MB. That's what I call small and beautiful. I've promised to keep all data stored locally, though, so more conversations and more images will inevitably take up some space — nothing I can do about that one.
Link: https://ds.imikufans.com/