Run DeepSeek Harness on Kimi K2 Instruct

by DeepSeekCoding agent

DeepSeek Harness (dsh) is DeepSeek's open-source agent harness, released under MIT and built so that everything is a plugin, from the model adapter and the tool registry to the agent loop itself. Start it with npx @deepseek-ai/dsh web and it opens a local web UI where the agent reads and edits your workspace, runs commands, and keeps a plan, with a headless profile for one-shot jobs from the terminal. On Opper it runs on any of 700+ models rather than DeepSeek's own alone: add Opper as a custom provider and you get one key, automatic fallbacks, and a choice of region.

Run

DeepSeek Harnessonin

Connect DeepSeek Harness to Kimi K2 Instruct

Set up with agents

Paste this into your coding agent (Claude Code, Cursor, Codex, and more) and it will set up and build with Opper for you.

Get your API key

Where you can run it

Kimi K2 Instruct runs on the routes below. Route only to EU providers when your data has to stay in Europe.

Provider	Region	Zero data retention	Training	Input	Output	Cache read
Novita	US		No	$0.57	$2.30	Input price

The model

Kimi K2 Instruct

Kimi K2 Instruct by Moonshot: post-trained 1T-parameter MoE for chat, instruction following, and agentic tool use with a 128K context window.

Available on

Pricing from$0.57/ 1M input$2.30/ 1M output131Kcontext

Privacy

ZDR
Training: No
Region: US

Run DeepSeek Harness on other models

Step-by-step setup for running DeepSeek Harness on each of these models through Opper.

More apps on Opper

Claude Code

Anthropic's terminal coding agent. Run it on any model on Opper, not just Claude.

Codingopper launch claude

Codex

OpenAI's open-source coding CLI. Run it on any model on Opper.

Codingopper launch codex

GitHub Copilot CLI

GitHub's terminal coding agent. Point its provider variables at Opper to run it on any model, no Copilot plan required.

CodingOpenAI-compatible