Tabbit
ResourcesBlogModels
Tabbit LogoTabbit

Tabbit — The AI Browser that Works for You

Topics

  • AI Browser Resources
  • Agentic Browser Resources
  • Browser Downloads and Install Guides
  • Browser Comparisons
  • AI Browser Alternatives
  • Browser Productivity Resources

Popular Guides

  • AI Browser
  • Agentic Browser Download
  • Best AI Browser 2026: Top 9 Tested & Ranked
  • AI Browser Download
  • Free AI Browser
  • Best AI Browser 2026
  • AI Browser Comparison 2026
  • AI Browser for Windows
  • AI Browser for Mac
  • Chrome Alternative 2026

Events

  • Tabbit Skill Competition
  • KPOP SBTI Fandom Personality Test
  • Tabbit Campus Creator Program
  • fifi's Picks: AI Skills for Research Papers
  • User Survey

About

  • Tabbit Blog
  • Press & Media
Prompt guide
OfficialDeepSeek V4 Pro

DeepSeek Harness v0.1: Launching and Configuring the Official Plugin-Based Agent Framework

Original source

GitHub (the official deepseek-ai/deepseek-harness repository) and the official X account @deepseekai

AuthorDeepSeek (official)

Source date2026-08-13

Tabbit curation2026-09-08

Read original

One-sentence takeaway

DeepSeek’s official open-source Agent framework is built on the principle that everything is a plugin (models, tools, skills, sessions, sandboxes, file systems, loops, orchestration, and UI can all be replaced); a single npx @deepseek-ai/dsh web command launches the Web UI locally. It is currently in developer preview and will introduce compatibility-breaking changes.

Use cases

  • Good fit: Building or customizing an Agent harness with DeepSeek’s official framework; experiments that combine models, tools, and skills as plugins; and studying the official default configuration for Agent workflows in the V4 series.

  • Poor fit: Directly relying on an unpinned preview framework in production; or needing a mature, ready-to-use Agent product (the preview iterates quickly and may break compatibility).

  • Applicable model version: The framework targets DeepSeek models (the V4 series), and the README uses the default Web UI launch as its example; the model plugin can be replaced.

  • Applicable client, Agent, or API: The dsh CLI/Web UI (default: http://127.0.0.1:3080); powered under the hood by the Cordis meta-framework.

  • Recommended reasoning tier and parameters: The framework documentation does not provide effort recommendations in the README; on the model side, continue using the official low/high/max configurations.

Directly usable content

The official README’s launch commands (original text):

# Run from npm (launches the Web UI at http://127.0.0.1:3080 by default and opens the browser)
npx @deepseek-ai/dsh web

# Do not open the browser automatically
npx @deepseek-ai/dsh web --no-open

# Run from source
git clone https://github.com/deepseek-ai/deepseek-harness.git
cd deepseek-harness
pnpm install
pnpm run build
pnpm dsh web

Key points from the official announcement (X @deepseek_ai, 2026-08-13):

DeepSeek Harness v0.1 is now available in developer preview!
It is open to developers of Agent frameworks worldwide, and the code repository is open-sourced under the MIT license.
Powered by the Cordis meta-framework, its core principle is: everything is a plugin.
Models, tools, skills, sessions, sandboxes, file systems, loops, orchestration, and UI can all be replaced.

Testing / Workflow Steps

  1. First confirm the Node.js environment, then launch the Web UI with npx @deepseek-ai/dsh web; in an SSH scenario, only the host URL is printed (local port forwarding is handled by the SSH client/editor).

  2. Read the development guide and architecture documentation in docs/ to understand the plugin boundaries (models/tools/skills/sessions/sandboxes/file systems/loops/orchestration/UI).

  3. Run a minimal Agent task using the examples in examples/, replace the default model plugin with deepseek-v4-pro, and record how the tool and skill plugins are loaded.

  4. Because this is a developer preview, pin the version and back up the configuration before upgrading; monitor the plugin ecosystem in GitHub Discussions and topics related to dsh-plugin.

  5. Compare the harness’s default behavior with direct API access on the target task (system prompt, tool protocol, and cost), then decide whether to use it in production.

Original evidence and data

  • Repository scale at collection time: approximately 12,940 commits, 176k stars, and 19.1k forks, under the MIT license; latest release 0.1.0-rc.8; includes directories such as docs/, examples/, packages/, python/, and apps/web.

  • The README explicitly says: “DeepSeek Harness is currently in developer preview and is iterating rapidly. THERE WILL BE COMPATIBILITY-BREAKING CHANGES.”

  • The README explicitly describes the plugin architecture as follows: “everything is a plugin, and is powered by Cordis, whose design is described in A Programming Paradigm for Spatiotemporal Composability.”

  • The official X announcement (2026-08-13) also announced the V4-Pro GA release, indicating that Harness and V4-Pro launched at the same time and that Harness is positioned as the official Agent framework.

Limitations of applicability

  • Developer preview: compatibility-breaking changes are frequent, as the README explicitly warns; this article records only the README and official announcement and did not install or run the framework.

  • “Everything is a plugin” is an architectural commitment, not a guarantee that every plugin is stable; third-party plugins must be audited independently (the MIT license does not guarantee security).

  • Harness’s Web UI/CLI is an official open-source DeepSeek project and is separate from DeepSeek’s commercial API service; take care with key management and costs when connecting an API key.

  • Framework capabilities and V4-Pro model capabilities are separate layers: the harness introduces its own system prompt and tool protocol, so “harness performance” must not be equated with “model performance.”

Source excerpt or observation (for compliant short quotation only)

The official X announcement says: “DeepSeek Harness is an Agent framework built around one core principle: everything is a plugin. Models, tools, skills, sessions, sandboxes, file systems, loops, orchestration, and UI …”.

Curated by Tabbit

Prompt material is summarized from public sources and Tabbit editorial notes. Check the original licensing and intended use before copying it.

DeepSeek V4 Pro

Use in Tabbit

DeepSeek V4 Pro

Related prompts

OfficialDeepSeek API Docs

DeepSeek-V4-Pro Thinking Levels and Tool-Calling Workflow

OfficialDeepSeek API Docs

DeepSeek-V4-Pro Responses Configuration Workflow in Codex

OfficialDeepSeek API Docs (Agent Integrations / Claude Code)2026-08-13

DeepSeek-V4-Pro 1M Context Environment Variable Configuration Workflow in Claude Code

MediaXSCT Bench (a scenario-based model selection and evaluation platform)

XSCT Bench “Autonomous Planning and Execution” Case: Agent Tool-Calling Prompt and Generated Result for deepseek-v4-pro

DeepSeek V4 Pro

Related reviews

OfficialDeepSeek API Docs2026-08-13

DeepSeek-V4-Pro Official Release: Reasoning and Agent Upgrades

MediaMindStudio2026-08-13

DeepSeek-V4-Pro-0813: MindStudio's Eight-Task Coding and Agent Hands-on Comparison

MediaReuters2026-08-13

Reuters DeepSeek-V4-Pro-0813: Official Pricing vs. Independent Index

MediaXSCT Bench (xsctbench.com, scenario-based model-selection evaluation)

DeepSeek-V4-Pro XSCT Bench Two-Case Comparison: Strong Planning, Weak Clarification