DeepSeek Harness

The model is the engine. The harness is the car. This is the part that decides whether you can drive it.

What the raw API gives you

A single endpoint: send messages, get a message back. That is enough for chat, and nothing else. Anything with tools, files, memory or multiple steps has to be built around it.

What the harness adds

Capability	Why it matters
Tool calling loop	Lets the model act, not just answer
File and workspace access	Read and edit real code, not pasted snippets
Session persistence	Long tasks survive a restart
Plugin system	Add capabilities without forking the harness
CLI and web UI	Two ways in, one engine underneath

What it does not do

It does not make the model smarter — a bad prompt is still a bad prompt
It does not manage your spend — that is your job, see the calculator
It does not replace infrastructure — you still host it

Why the licence matters

DSH is MIT licensed. That means you can read the source, modify it, and ship it commercially without opening your own changes. For teams deciding whether to build on it, this removes the usual "what happens if they change the licence" risk.

Agent token costs escalate

A harness makes agents possible, and agents are expensive. In our agent cost analysis, a task adding 3,000 tokens of context per turn reaches 20x the first turn's input by turn 20. Budget for it before shipping.