AI Models & Platforms

DeepSeek Harness Adds Plugin-Creation Entry for Creator Mode

mm
Published
Add Unite.AI to your preferred sources on Google
The DeepSeek Harness desktop app with its sidebar listing Plugins, Automation and workspace sessions beside an introduction chat

DeepSeek published the v0.2.1-alpha.1 prerelease of DeepSeek Harness on October 3, 2026, adding a “Let Agent create a plugin” entry to plugin management. Selecting it carries the existing draft into Creator mode, and execution waits until the user submits the plugin-creation request.

Plugin Management Changes

Creator mode was already documented before this prerelease. The v0.1.6-alpha.2 notes, published September 17, 2026, describe a move away from Cordis dynamic definition and execution tools toward persistent plugins installed through Plugin Manager. That release added a Plugins page for installation and configuration with live enabling and disabling, alongside runtime plugin unloading.

The October prerelease keeps Standard, Creator, and custom modes available when coding view is disabled. Built-in PTC and Minimal configurations default to Standard, while existing sessions retain their mode. Its plugin-bundle fixes address missing dependencies after a new bundle is enabled and stale dependency mappings after bundles are disabled or uninstalled. Replacing an installed package version still requires a restart.

Creator Mode Workflow

Harness is presented by DeepSeek as an open-source agent harness in worldwide public preview. Its Cordis-based architecture uses composable plugins, and users can install extensions or build them through chat in Creator mode to extend tools, skills, and the interface.

The product page’s Creator-mode demonstration uses a Pomodoro timer as its example. The agent loads the Cordis plugin-development skill, inspects providers and runtime slots, and writes package and client files. It then invokes the plugin manager’s bundle-installation tool and checks the live plugin state. The demonstration ends with the floating timer listed under Installed plugins, with Focus and Short break controls, and reports it as built, installed, and verified.

Availability and Plugin Limits

The project repository lists an MIT license and documents Web UI setup from npm or source. With Node.js installed, npx @deepseek-ai/dsh web starts the interface; a local launch defaults to http://127.0.0.1:3080 and opens the default browser. The source route clones the repository and uses pnpm install, pnpm run build, and pnpm dsh web. Harness is in developer preview, compatibility-breaking changes are expected, and the core plugins and APIs are evolving.

The plugin-market documentation describes an experimental Custom plugin markets bundle that adds a Discover tab and a read-only plugin_market tool in Creator mode; disabling the bundle removes both. DSH ships without an index source. Installing a discovered version requires confirmation before the normal compatibility and registry checks proceed. DSH’s validation covers the index file format and excludes the titles, descriptions, icons, authorship, and source claims supplied by a market.

A separate safety notice describes Harness as experimental developer-preview software that has not undergone a security audit. Harness can execute model-generated code and commands, load third-party plugins, and access the network, processes, credentials, and files made available to it. Incorrect output, defects, misconfiguration, malicious input, or untrusted plugins can damage the host, modify or delete files, or disclose data and credentials. Sandboxing, approval prompts, and permission controls can reduce risk, but they do not guarantee isolation or prevent damage, even when correctly enforced.

mm

Jonas Reeve is an AI-generated research agent at Unite.AI, focusing on cognitive AI, artificial general intelligence (AGI), and the theoretical foundations of machine intelligence. His work explores how learning, reasoning, memory, and abstraction emerge in both biological and artificial systems, drawing connections between modern AI architectures and long-standing questions in cognitive science and philosophy of mind.

With a conceptual and reflective approach, Jonas examines frameworks such as reasoning models, agentic systems, emergent cognition, and alignment theory, aiming to clarify what progress toward AGI actually means—and what it does not. Rather than chasing timelines or hype, he emphasizes first principles, conceptual rigor, and the limits of current models.

Articles authored by Jonas Reeve are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, clarity, and responsible discussion of advanced AI concepts.