SI Agent News
DevelopingOpen models

dsh-purge adds local controls for DeepSeek Harness plugins

A new community add-on provides a dashboard and command-line tools to manage, backup, and roll back configurations for self-hosted DeepSeek Harness setups.

Blueprint drawing: picture for the story 'dsh-purge adds local controls for DeepSeek Harness plugins'
AI-made blueprint drawing for this story. No real people or products are shown.

The short version

Why it matters

For users who self-host AI tools, managing complex plugin configurations can be risky because one wrong change can break the setup. This tool adds a safety net by providing a way to test changes and undo them instantly.

Managing complex AI setups

Managing a self-hosted DeepSeek Harness setup often requires constant adjustments to plugins, prompts, and host patches. Users frequently face the challenge of knowing exactly what state their system is in after making multiple modifications. The dsh-purge Community project aims to solve this by providing a centralized way to monitor and control these variables.

As news.lavx.hu reported, the project focuses on reducing interaction friction for those who use the Harness for learning or personal projects. It provides a way to see evidence of whether specific patches have actually taken effect. This transparency helps users decide which changes to apply without guessing the outcome.

The tool is designed as a local add-on, meaning it runs on the user's machine. It does not alter the underlying AI model itself or send any prompts or configurations to a cloud account. This keeps the user's data and local environment private while providing a more organized management experience.

How the control panel works

The dsh-purge Community tool combines a visual sidebar panel with a suite of command-line tools. The panel organizes secondary controls into collapsible areas to keep the interface clean while placing the main apply action at the top. It includes a maintenance page where users can view recent actions and monitor backup storage usage.

The command-line tools offer deeper diagnostic capabilities, such as reporting host paths and counting files. They can also diagnose patch-anchor drift, which identifies where a patch might have moved within the files. Furthermore, a specific probe can check whether a runtime injection actually reached the model successfully.

One of the key features is the ability to take snapshots before performing selected operations. The project preserves original files and a verification record when a snapshot is created. This allows users to experiment with new host patches or plugin updates while maintaining a reliable way to return to a previous state.

Prompt and rule management

Users can manage their prompt presets, rules, and reinforcement segments directly through the new interface. The tool allows for specific exclusions, such as leaving certain commands or panel tools available while excluding prompt injections for a selected preset. This level of granularity helps users tailor their assistant's behavior more precisely.

For those importing new skills, the plugin supports several archive types, including ZIP, TAR, and TGZ. It validates the paths and extraction size before installing the contents. This automated validation helps prevent errors during the setup process for complex new capabilities.

The project also separates the installation of a plugin from the application of host patches. This distinction is important because it allows users to install software without immediately modifying the host's core behavior. Users can choose when to apply patches and when to restart the host independently.

Who benefits from these tools

The project is specifically designed for learners, personal project maintainers, and those running automation experiments. A learner can use the status checks to verify that a plugin loaded correctly before continuing an experiment. This prevents the frustration of troubleshooting a broken configuration that was never actually active.

For those maintaining a personal setup, the tool provides a consistent place to manage presets rather than visiting separate controls during every session. It is particularly useful for users who want a recoverable workflow, as they can save local overrides for confirmed patch anchors. If a change causes trouble, they can roll back the plugin snapshot immediately.

The developers emphasize that the tool is intended for environments the user owns or has permission to use. It is framed as a tool for personal development and authorized maintenance rather than a way to modify systems without consent. This ensures the tool remains a helpful utility for legitimate administrative tasks.

Technical limits and roadmap

While the tool provides extensive local controls, it has some specific compatibility limits. The project lists tested host versions, but other hosts may still install with an unverified status. In these cases, users are encouraged to manually inspect the diagnostics to ensure the setup is functioning as intended.

The tool's mobile capabilities are currently limited to read-only functions on an Android emulator. The developers have not validated write operations, such as applying patches or rolling back snapshots, on real mobile devices. This means the full suite of recovery tools is primarily intended for desktop and server environments.

Recent updates have focused on making the interface more consistent across different views. The maintainers have unified the settings components so that update intervals and package details appear the same everywhere. These changes ensure that as the project grows, the user experience remains predictable and organized.

What to watch

Try it

If you use DeepSeek Harness, check the GitHub repository for dsh-purge to see if your host version is on the tested list.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by news.lavx.hu · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news