Should a personal agent hide its execution—or let you inspect it?

When I started building DeepSeekBot, I wanted the convenience of talking to a personal bot without losing the visibility of working directly with a coding agent.

That led to a design choice: Bot mode should be a different way to interact with DeepSeek Harness, not a separate execution system that replaces it.

DeepSeekBot is a plugin for DeepSeek Harness. Bot mode adds a bot-first interface, while the bot’s agent execution uses underlying Harness Sessions. You can open an available Session from the bot’s Sessions list and return to the native Harness interface to inspect its conversation and execution history.

Here is the workflow I want to make easy:

1. Give your personal bot a task, such as investigating a small coding issue.

2. Look at the Sessions associated with that bot, including its orchestration and task execution Sessions.

3. Open the relevant Session in the native Harness interface to understand what happened, rather than relying only on the bot’s final summary.

For me, the benefit is not “a bot that never makes mistakes.” It is a bot whose work is easier to understand and troubleshoot.

A conversational interface is useful when you want to delegate. An execution-oriented interface is useful when you need to inspect. I do not think users should have to choose between them.

The shorthand is “GrokBot-style convenience meets a coding-agent workflow.” Think of the direction as GrokBot + Cursor—not a claim that we reproduce every feature of either product.

Building on DeepSeek Harness also means we are not trying to develop every capability ourselves. Its open-source plugin ecosystem gives us a foundation for extending what bots can do through compatible plugins. Compatibility still matters; this is not a promise that every plugin combination works unchanged.

Persistent memory and personal bot teams are part of the broader vision. But the principle behind Bot mode is simpler: convenience should not require hiding the execution underneath.

When you delegate work to a personal agent, how much of the underlying execution do you want to see—and what would you want to inspect first?

Project and installation information: https://deepseekbot.app