Superpower Daily
Broadcom Validates Five AI Model Families for VMware’s On-Premi...Nvidia’s $10B Physical-AI Business Bets on a Full Stack—and an...John Ternus Takes Apple’s Top Job as AI and China Pressures Con...Claims Adjusters’ AI Backlash Hits 98% Negative Reviews as Insu...Replica Cyber Ships Isolated AI Agents With Runtime Controls fo...Hack The Box Benchmark: Best Human Team Cleared 36 Challenges;...Broadcom Validates Five AI Model Families for VMware’s On-Premi...Nvidia’s $10B Physical-AI Business Bets on a Full Stack—and an...John Ternus Takes Apple’s Top Job as AI and China Pressures Con...Claims Adjusters’ AI Backlash Hits 98% Negative Reviews as Insu...Replica Cyber Ships Isolated AI Agents With Runtime Controls fo...Hack The Box Benchmark: Best Human Team Cleared 36 Challenges;...
Toolspublished

Modellix Adds 28+ Models, Including $0 Options, to DeepSeek Harness Beta

The beta gives users of the rapidly adopted coding agent one model-selection layer and web-tool routing, while its free-call pitch still requires a Modellix account and funded balance to create an API key.

By 3 min read
Modellix Adds 28+ Models, Including $0 Options, to DeepSeek Harness Beta
Modellix Adds 28+ Models, Including $0 Options, to DeepSeek Harness Beta
Sign in to save and react

Listen to this story

The audio brief

About 1:35
0:001:35
Read transcript
DeepSeek Harness can now reach more than 28 models through a new Modellix beta plugin, including two options billed at zero dollars per call. The plugin, called dsh-modellix, puts Modellix’s multi-provider catalog inside Harness’s native model selector, so developers can switch among providers without replacing the API key used by the agent. That catalog includes OpenAI, Anthropic, Google, xAI, GLM, Qwen, DeepSeek, and Kimi models through one gateway. The timing matters because Aurora Mobile says DeepSeek Harness accumulated nearly 200,000 GitHub stars soon after its open-source debut, giving integrations like this a potentially large distribution point. The free-call offer does have an important condition. Users still need a Modellix account, an API key, and a topped-up account balance before that key can be created. The two named zero-cost choices are modellix-ai/free-llm, which routes requests to current open-source models rather than selecting one fixed model, and zai/glm-4.7-flash. The LLM layer supports OpenAI-compatible Chat Completions and Responses APIs, along with Anthropic-compatible Messages APIs. Web tools can also route through Modellix, while Design features for image, video, and audio generation remain in preview. The beta can be installed with the dsh plugin command, but its behavior and APIs may change. The key thing to watch is whether this broad catalog and web routing prove reliable enough for daily coding work as the integration matures.

Story brief

3 key points

Aurora Mobile’s dsh-modellix beta turns DeepSeek Harness into a gateway to Modellix’s multi-provider catalog, letting developers switch models without replacing the agent’s API key. The plugin exposes more than 28 models, but its two $0-per-call options still require a Modellix account and prepaid balance to create an API key. LLM support is functional, while Design remains in preview and the beta’s APIs may change....

  1. 01

    The catalog spans OpenAI, Anthropic, Google, xAI, GLM, Qwen, DeepSeek and Kimi through one gateway key.

  2. 02

    modellix-ai/free-llm routes requests to current open-source models; it is not a fixed model selection.

  3. 03

    The plugin supports OpenAI Chat Completions and Responses APIs plus Anthropic Messages APIs.

Aurora Mobile’s Modellix has released dsh-modellix, a beta plugin that brings its LLM gateway into DeepSeek Harness’s native model selector. The immediate draw is access to more than 28 models through one API key, including two models billed at $0 per call, for a coding agent that Aurora Mobile says reached nearly 200,000 GitHub stars shortly after its open-source debut.

A gateway inside the agent

DeepSeek Harness was open-sourced on August 13. Aurora Mobile said it had accumulated nearly 200,000 GitHub stars by August 27, a rapid uptake that gives third-party plugins a potentially large distribution point. Harness is built around plugins, and dsh-modellix uses that structure to place Modellix’s catalog directly in the agent’s model picker.

The plugin’s LLM, Design and Web functions can be toggled independently. Its working LLM layer synchronizes Modellix’s live catalog, spanning providers including OpenAI, Anthropic, Google, xAI, GLM, Qwen, DeepSeek and Kimi. That means a developer can change among models without changing the API key used by the agent.

The zero-cost offer has an entry condition

The two $0-billed choices are modellix-ai/free-llm, which Aurora Mobile says automatically routes requests to current open-source LLMs, and zai/glm-4.7-flash. The first is a routing label rather than a fixed model named in the release; developers wanting a particular model can select from the larger catalog, where pricing varies.

The offer is not a sign-up-free path. Users need a Modellix account and API key, and Aurora Mobile says account balance must be topped up before an API key can be created. The company nevertheless states that calls to the two named models are billed at $0, separating per-call pricing from the account-setup requirement.

One integration, uneven maturity

The model side supports OpenAI-compatible Chat Completions and Responses APIs, plus Anthropic-compatible Messages APIs. Modellix also says compatible clients can redirect requests to its gateway with a base-URL override instead of a code rewrite. For teams already using those API styles, the plugin is one route to consolidate model access inside Harness.

The broader feature set is not equally ready. Web mode routes Harness’s web_search and web_fetch tools through a Modellix endpoint. The LLM capability is functional in this beta, but Design—intended for image, video and audio generation—remains in preview while Modellix works across upstream model schemas. Aurora Mobile also says the beta’s APIs and behavior may change before a full release.

Install path
dsh plugin --profile web add dsh-modellix

That leaves the practical test for users: whether the gateway’s catalog, web routing and model compatibility are dependable enough for daily coding work as the plugin evolves. For now, dsh-modellix is a beta access layer, not a finished multimodal workspace.

Sources

  1. stocktitan.netDeepSeek Harness Draws Nearly 200,000 GitHub Stars as Aurora Mobile Adds Free AI Models
  2. manilatimes.netAurora Mobile’s Modellix Releases Beta Plugin for DeepSeek Harness, Adding Free LLM Models to the Fast-Growing Open-Source Coding Agent

Related coverage

Continue reading