Introduction

The design philosophy of DeepSeek Harness is “everything is a plugin,” aiming to let developers extend its capabilities through plugins. When developing or debugging with the web frontend chat.deepseek.com, historical conversations are often scattered across the browser and difficult to reuse or organize. How can these conversations be safely saved locally and precisely imported into a DSH workspace as a new Agent session for continued development? The Convoport plugin solves this process gap: it automatically captures web conversations, stores them locally, and provides message-level import functionality.

Plugin Positioning

  • Name: Convoport
  • Maintainer: MrElysium
  • One-liner: Capture AI web conversations into local JSON + SQLite, and import selected messages into agent harnesses as resumable sessions.

Core Features

  1. Automatic capture: Monitors chat.deepseek.com through a browser extension (MV3), uses the official API to pull conversations, and automatically synchronizes them to the plugin.
  2. Local storage: Data is stored in the $DSH_HOME/capture/ directory, saved in JSON format, with one file per conversation. The data is fully localized and is not uploaded through any server.
  3. Token savings statistics: Provides cumulative saved token count, conversation count, captures today, and estimated money savings (calculated based on a reference price).
  4. Message-level import: Supports selecting specific messages before import, with quick-selection options (such as “first 3 / last 3”). After import, the selected content becomes the history of a new Harness session.

Installation and Activation

Installing the plugin requires two steps: installing the plugin itself via DSH, and manually loading the browser extension (required by Chrome’s security restrictions).

  1. Install the plugin itself:
    dsh plugin --profile web add convoport

After installation, restart dsh web and refresh the browser.
  1. Load the browser extension:
    The extension files are located in the plugin’s extension/ directory (they can be found in the npm package under node_modules/convoport/extension/ or at the repository root).
    • Open the Chrome extension management page: chrome://extensions.
    • Enable “Developer mode” in the upper-right corner.
    • Click “Load unpacked extension” and select the extension/ directory.
    • (Optional) Click the extension icon to open Options, and confirm that the backend URL is http://127.0.0.1:3080 (the dsh web port).

Typical Usage

After enabling the extension, normal conversations on chat.deepseek.com are captured automatically.

  1. View status and list: A “⛁ Convoport” button appears at the bottom of the DSH Web sidebar. Click it to view the statistics card (total tokens, conversation count, captures today, estimated savings) and the capture list.
  2. Select messages: Click a conversation in the capture list to open the message selection interface. You can manually select the messages you want to keep, or use the quick-select buttons (such as “first 3 / last 3 / all / clear”).
  3. Create a new session: In the message selection interface, click “Import selected → new session (N)”. The plugin creates a new Harness session in the current workspace and uses the selected messages as the historical context for that session.
  4. View statistics: Enter the /capture command in any DSH session to quickly view a statistical overview.

Notes

  • Manual loading limitation: Due to Chrome’s security policy, unpacked extensions cannot be loaded directly from the command line; you must manually perform “Load unpacked”.
  • Data location: All captured JSON data is stored in the $DSH_HOME/capture/ directory. The content is plain text and can be read by any developer.
  • Token counting logic: The plugin prioritize uses DeepSeek’s official accumulated_token_count field. If this field is not provided, it falls back to a heuristic (4 characters/token + structural overhead).
  • Pricing reference: Estimated money savings are calculated based on a reference price of $0.02 / 1k tokens.

Conclusion

Convoport focuses on converting web-based conversation assets into local Sessions directly usable by DSH. It does not alter the core architecture; instead, it acts as a connector, allowing developers to manage historical conversations freely and continue development seamlessly in a local environment.