Introduction¶
When running tasks in DeepSeek Harness (DSH), tasks may fail silently or hang in the background until you check the console. DeepCanary is a client plugin that creates a local task inbox, aggregates issues, approvals, and completion summaries, and sends quiet notifications when you need them.
Plugin Positioning¶
DeepCanary is a local client plugin for DeepSeek Harness, used to manage task reminders, the inbox, and session resumption. It follows the DSH “everything is a plugin” philosophy and addresses delayed information delivery for long-running tasks through a local inbox mechanism. The plugin is maintained by Oscar-Williams and is licensed under the MIT License.
Core Features¶
- Local task inbox: Aggregates issues, approvals, failures, and completion summaries into a local inbox, providing a unified entry point for viewing.
- Notification tier configuration: Supports three notification levels—C1 (check recommended), C2 (requires attention), and C3 (immediate attention)—and allows configuring hourly budgets and quiet hours.
- Quick session return: Clicking an inbox item can directly return you to the related session for reviewing evidence or providing feedback.
- Auxiliary operations: Supports keyboard controls, the stash feature, and browser notifications.
Installation and Enablement¶
Install the specified version through the DSH plugin manager:
dsh-deepcanary@0.1.4-rc.1
After installation, open DeepCanary in the DSH sidebar, go to Settings, configure notification levels, budgets, and quiet hours, and enable browser notifications to start using it.
Notes¶
- Runtime dependencies: The notification feature requires both the host and renderer processes to remain running, and browser notification permissions must be granted.
- Platform limitations: The Windows WebUI is the current development and device verification entry point. The hidden-window wake-up feature for desktop versions requires public host support and device evidence support (the current RC version provides the renderer notification path).
- Experimental features: Supervisor is experimental and disabled by default.
- Privacy protection: Reminder state contains only metadata, hash references, and session handles. It does not retain conversation text, prompts, model outputs, tool parameters, or API credentials. Feedback collection is optional.
Summary¶
DeepCanary provides a localized entry point for task management, suitable for users who need to run tasks in the background for long periods and want timely updates on key milestones. Its core value is centralizing scattered task states into an inbox and reducing interruptions through a tiered notification strategy.