Fish and sheep, from Aofeisi Temple
QuantumBit | Official Account QbitAI
The DeepSeek official Muse hasn’t arrived yet, but the underground experts are already eager.
The newly served item is “DeepSeek-Bot”:
Robot team based on DeepSeek Harness.
△Image source: Author @Fei2411
Like Grok Bot, the ds bot has developed a multi-Bot team interface based on DeepSeek Harness.
Each Bot can have its own name, role, and model. More importantly, it has “persistence” – the ability to accumulate long-term memory.
Additionally, to save tokens, you can also assign a supervisor to the Bots. Contact the supervisor in advance, and then let it coordinate with the Bots.
Author FeiZ also marked the key points:
You can integrate any model you want into the ds bot. In the future, you might even be able to integrate local Agents.
ds bot inherits the “everything is a plugin” concept of dsh to maintain extensibility.
Although unofficial, it seems this is the authentic flavor of a big blue fish (bushi)?
In less than a day since it went live, the Big Fat Fish enthusiasts have given this project over 250 stars.
I can’t afford to keep the Grok Bot, why not give the ds bot a try to see how it tastes?
DSH-based bot plugin
Everything is a plugin, and the ds bot is also in the form of a plugin.
The main functions are as follows.
If there are problems, contact the Bot supervisor first
In the ds bot, each Bot has an independent identity, and you can set its name, role description, commands, image, as well as the model called behind it.
The bots have a persistent operating state and possess long-term memory.
You can also set up a Main Bot. This Bot manager has high permissions and can help you manage and configure other Bots and group chats. It not only reduces the complexity of managing multiple Agents, but also lowers token consumption in an environment where group chats consume tokens rapidly.
In this regard, ds bot has been specially optimized: when setting an administrator in a group chat, your messages will be replied by the administrator by default, and it will help you call other bots.
Of course, you can also directly @ the bot that needs separate communication to have a one-on-one conversation.
△Image source: Author @Fei2411
In summary, the author strongly recommends configuring other Bots and group chats directly through conversation with the Bot.
By the way, the author also mentioned that the Bot’s model can be switched at any time. However:
I still recommend using the same model for long-term use, and only switching it when it is truly necessary.
Long-term memory
Long-term memory is a key feature of the ds bot. Memory can be managed on the details page, or it can be handled directly through dialogue management.
Technically, each Bot corresponds to a DeepSeek Harness Session at the underlying level, having its own working context.
When encountering long conversations that exceed the context window, the ds bot uses mechanisms such as compression, summary transfer, and local search to manage the context.
ds bot has also set up memory storage based on Markdown files for each Bot, and designed a team-sharing memory system.
Additionally, after each round of conversation, the model will undergo an automatic memory review to determine which information needs to be saved, updated, or deleted.
Other technical features include:
Usage statistics: In Settings → Usage, you can view Token usage by time, Bot, and model.
Data localization: All data are stored in the bot/ and sessions/ folders under the DSH main folder on your computer. Uninstalling the plugin does not delete this data.
Open source and licensing: The project uses the Apache License 2.0.
Installation and Usage
Both the desktop and web versions of DSH allow use of ds bot.
On the desktop version, select “Plugins” in the sidebar → “Add Plugin”, and enter “ds-bot” to install it.
On the web side, Node.js 22.19 or later version, along with pnpm, are required. Then, add the plugin using the following command:
npx @deepseek-ai/dsh@0.2.1-alpha.1 plugin –profile web add ds-bot
npx @deepseek-ai/dsh@0.2.1-alpha.1 web
For more details, please refer to the project homepage.
So, if you're interested, you can start using it. Feel free to come back to the comment section and share your usage feedback~
Project address:
https://github.com/FeiZhuLulu/DeepSeek-Bot/tree/main
— End —