Basic questions
What is DeepSeek Harness?
DeepSeek Harness (command dsh) is an open-source agent harness built by DeepSeek AI. It performs work on your computer — reading and changing files, running terminal commands, searching the web, creating documents — rather than only answering questions. All capabilities are installed as plugins, so it can be extended.
Is DeepSeek Harness free?
The software is free and open source (MIT licence). What costs money is model usage: a DeepSeek API key or another provider is billed per token. With a local model there is no API cost either, only your computer's resources.
Who is it for?
People whose work is file-based and repetitive: developers, marketing teams managing many pages, finance staff producing routine reports, researchers reading many sources, and business owners reducing admin work.
Is DeepSeek Harness the same as DeepSeek chat?
They differ. The chat app answers questions in conversation. DeepSeek Harness is a program running on your computer with limited permission to work on files and run commands. Both can use the same DeepSeek models.
Does DeepSeek Harness support Indonesian?
Yes. You can give instructions in Indonesian and request Indonesian output. For technical terms, adding the original term in brackets usually keeps results consistent.
Is there a desktop app?
Yes. DeepSeek Harness ships as a desktop app for macOS and Windows, alongside the local web interface and CLI usage. Pick what suits you: desktop for general users, local web for daily work, CLI for automation.
Technical questions
Minimum computer and OS requirements?
Node.js ^22.19.0 or >= 24.0.0, Git, and Linux x64/arm64, macOS 14+ (arm64) or Windows x64. Python 3.10+ is only needed for the Python SDK. Because models run on provider servers, an ordinary computer is enough.
Which models can be used?
DeepSeek models (including a fast variant for light tasks and stronger variants for heavy work), third-party providers such as Anthropic, OpenAI, Kimi and GLM, your own gateway using three supported protocols, and local models via Ollama. Model names and availability can change during developer preview.
Can it be used offline?
Partly. File operations, code search and the terminal run locally. Model calls still need a provider connection unless you run a local model. Web search also needs the internet.
How do we add company-specific capabilities?
Three options, easiest first: (1) a skill — a markdown file holding a procedure, (2) an MCP server to connect internal systems, (3) a TypeScript plugin when you need a genuinely new tool or UI. For most needs, MCP is enough.
Can it run several tasks at once?
Yes. Subagents handle self-contained tasks, JavaScript workflows coordinate dozens of agents in stages, and background jobs cover long commands. Note that parallel work consumes more tokens.
What if the output is wrong?
That is why version control matters: file changes can be compared and reverted. To reduce errors up front, ask for a plan first, ask the agent to verify itself (run tests, reopen output), and split work into reviewable stages.
Safety, privacy and cost
Is DeepSeek Harness safe?
Safety depends on your settings. Three sandbox modes limit file access and a human approval policy covers sensitive actions. Still, the software is developer preview and has not been through a comprehensive security audit, so do not give it unsupervised full access to production systems.
Which permission mode should I use?
Start with read-only while exploring. For daily work, workspace-write with ask approval is the most balanced choice. Use danger-full-access only when the task truly needs access outside the workspace and you understand the risk.
Is my data sent to the internet?
Conversation content and files the agent reads are sent to your chosen model provider for processing. If data must not leave your network, use a local model. Web search also sends outbound requests, so avoid searching for confidential information.
What does it cost per month?
There is no licence fee. Model cost is billed per token, so it depends heavily on volume. For light work a few times a day it is usually small; workflows with dozens of agents scale that cost up. Start small and measure usage.
Can it be used with customer data?
Only if your privacy rules allow it. Recommended practice: anonymise data before processing, use a local model or internal gateway for sensitive data, restrict the workspace, and keep an audit trail of the work.
Can I get hands-on help?
Yes, and the first consultation is free. Contact Andi at +62 897-6003-363 (phone or WhatsApp) to discuss your needs, the risks, and the most sensible first step.
Need help rolling out DeepSeek Harness?
Free consultation with Andi to choose the right setup, prepare a workflow, or review safety before your team adopts it.