Aurora Mobile Limited (NASDAQ: JG) announced that its Modellix.ai platform has released a beta plugin called dsh-modellix for DeepSeek Harness, an open-source AI coding agent that accumulated nearly 200,000 GitHub stars within two weeks of its August 13, 2026 debut.

The plugin connects Harness users to Modellix's large language model (LLM) gateway using a single API key. Two models are available at no per-call cost: "modellix-ai/free-llm," which routes requests to open-source LLMs, and "zai/glm-4.7-flash." Users can also access more than 28 models from providers including OpenAI, Anthropic, Google, xAI, and DeepSeek through the same key.

The plugin offers three independently toggled capabilities: LLM, Web, and Design. The LLM function syncs Modellix's model catalog into Harness's model selector. The Web function routes search and fetch tools through Modellix's endpoint. The Design function enables image, video, and audio generation, though it remains in preview. The LLM capability is described as functional in the current beta release.

Modellix's gateway is compatible with OpenAI and Anthropic protocols, allowing existing configurations to connect via a base-URL change without code rewrites.

"AI coding agents are becoming the default front door to AI development, and the ecosystem around DeepSeek Harness is growing remarkably fast," said Weidong Luo, Chairman and Chief Executive Officer of Aurora Mobile. "dsh-modellix is an early beta, and our goal is simple: a developer using Harness should be able to try the latest models — including those billed at $0 — in under two minutes."

The beta plugin, version 0.1.1, is available through DeepSeek Harness's plugin command. Users are required to create a Modellix account and top up an account balance before generating an API key, though the free-tier models carry no per-call charge.