A community-built plugin called DS Bot is giving users of DeepSeek Harness, a tool that runs AI models, a new way to work with artificial intelligence. According to the project's page on GitHub, DS Bot turns the harness into a set of long-lived bots that users create, talk to and assign work to. The plugin also supports group chats, where several bots can collaborate, and it lets bots hand tasks to one another.
The project is open source and released under the Apache License 2.0. Its maintainers describe it as an independent community effort rather than an official product of DeepSeek.
What DS Bot does
DS Bot is designed around a simple idea: instead of starting a fresh conversation every time, users build their own bots and keep working with them over time. Each bot has its own details page, where users can see its information and choose which model it uses. The source material notes that DS Bot relies on the models already set up in DeepSeek Harness, so users need at least one model configured before they begin.
Memory is another central feature. According to the project documentation, a bot remembers what a user likes and applies that knowledge to later tasks. Users can review or edit this memory from the bot's details page, or simply tell the bot what to remember.
A main bot as the starting point
When users first open DS Bot, they find a main bot named Chief already in place. The main bot holds the highest level of permissions in the system. It can manage the other bots and the group chats. The project recommends that new users treat it as the first point of contact, since it walks them through the setup process. In the project's own summary, the idea is simple: if you have a question, ask the main bot.
Group chats and bot-to-bot work
Group chats are one of the plugin's most distinctive features, and also one of its most expensive to run. The project itself acknowledges that multi-bot conversations cost more, so DS Bot assigns every group chat an admin. By default, the admin answers messages and brings in other bots when needed. Users can also address a specific bot directly by typing an at sign and choosing it from a list.
Bots can pass work to each other in the same way. When one bot asks another for help, the second bot completes the task independently while the rest of the group continues. The group, the sidebar and the bot's own chat window all show progress. Once the work is finished, the result is posted to the group, and the bot that made the request responds to it. If a user wants to halt the work, they can do so from the group's details page.
Users can also give a group chat access to another conversation as background context, using the reference option in the message box. This allows bots to draw on earlier discussions without the user repeating themselves.
Installing the plugin
The installation process begins with DeepSeek Harness itself. Users must install it and configure at least one model. From there, they open the sidebar, select Plugins and then Add plugin, and enter ds-bot as the package name. Users in mainland China can switch the registry to a mirror located there. After installation, the plugin must be enabled, and in some cases the user must completely quit and reopen the harness for the change to take effect.
If the standard package source cannot be reached, the project lists alternative routes. Users can download a release file from the project's Releases page and enter its full path, or install directly from the GitHub repository, which requires Git on the computer. These options give users some flexibility when network access is limited.
Capabilities, usage and data control
DS Bot includes a shared library called Capabilities, where users can store MCP servers, skills and packs. A capability added to one bot becomes available from that bot's next message, both in its own chat and in group chats. The project includes a safeguard: any local MCP command added by a bot requires the user's approval before it can run. While other bots are using a capability, only the user or the main bot can change or delete it.
The plugin also offers a usage view that shows token consumption by time, by bot and by model. The project states that all data stays on the user's own computer, in specific folders within the harness home directory. Uninstalling the plugin does not remove that data, which may matter to users who want to keep their bot history.
The project's disclaimer deserves attention. Bots use the user's own model quota, and they can read and write files and run commands when asked to. The maintainers advise users to check important actions themselves before relying on them.
Outlook
DS Bot reflects a broader movement toward AI assistants that persist across sessions, remember preferences and coordinate with one another rather than answering isolated prompts. Its success will depend on whether the group chat model proves useful enough to justify its cost, and on how the community responds to the plugin's permission controls. For users already working inside DeepSeek Harness, the project is a concrete example of how far customizable bots can go today.
Related: Open-Source Port Brings Diablo II: Resurrected to iPhone and Mac · SolidCraft: Open-Source Parametric 3D CAD Built in Rust
Comments (0)