What Is the Modellix DeepSeek Harness Plugin?
DeepSeek Harness has exploded on GitHub with nearly 200,000 stars in two weeks — thanks to its “everything is a plugin” architecture that lets developers extend their AI coding workflow with the tools they already love. Today we’re introducing dsh-modellix (beta), the Modellix plugin for Harness. It brings our multi-provider LLM gateway — including models billed at $0 — directly into Harness’s native model selector, powered by a single Modellix API key.
Before we dive in: we build Modellix, a unified AI gateway for LLMs and media generation. We have a commercial interest in this announcement. Every capability described below is verified against our public documentation as of August 28, 2026.
Why this matters
Coding agents are fast becoming the front door to AI development. The models behind that door should be a choice, not a lock-in. With dsh-modellix, you can:
- Try LLMs at zero per-call cost. Switch your Harness conversation model to
modellix-ai/free-llm— billed at $0 — orzai/glm-4.7-flash, also billed at $0. Creating an API key requires topping up your balance first (min. $5). - Use 28+ models with one key. OpenAI, Anthropic, Google, xAI, GLM, Qwen, DeepSeek, Kimi — all behind the same gateway, no separate vendor accounts.
- Never edit a model ID again. Latest aliases (e.g.,
~openai/gpt-latest) automatically route to the current flagship of each series. - Keep your existing setup. The gateway speaks OpenAI-compatible Chat Completions & Responses plus Anthropic-compatible Messages. If your client already works with OpenAI or Anthropic, it works with Modellix.
What’s in the plugin
| Capability | What it does | Status |
|---|---|---|
| LLM | Syncs Modellix’s live model catalog into Harness’s model selector; switch models without reconfiguring | Available in beta |
| Web | Routes Harness’s native web_search and web_fetch through Modellix | Available in beta |
| Design | Image, video, and audio generation in Harness’s Design view | Preview |
We shipped LLM and Web first because they power everyday development. Design is in preview while we harden the experience; we’ll update this page when it’s production-ready. As a beta, the plugin’s APIs and behavior may change before full release.
Architecture at a glance
The dsh-modellix plugin registers three independently toggled capabilities. Harness calls the Modellix Gateway via OpenAI-compatible Chat/Responses and Anthropic-compatible Messages. The gateway then fans out to providers like OpenAI, Anthropic, Google, xAI, GLM, Qwen, DeepSeek, and Kimi — all with a single key and consistent observability.
Get started in two minutes
- Install —
dsh plugin --profile web add dsh-modellix, then restart your Harness Web profile. - Add your key — paste your Modellix API key in “Connect Modellix” or in the Modellix settings page. No key yet? Create an account at modellix.ai, then top up your balance (min. $5) to generate a key.
- Pick a model — open Harness’s model selector, choose the Modellix provider, and select
modellix-ai/free-llm. Start calling.
For reference, running the install step should produce output along these lines (example only — exact verbatim may vary by platform):
1 | $ dsh plugin --profile web add dsh-modellix |
After picking modellix-ai/free-llm in the model selector, your first request goes straight through the gateway — example session below:
1 | You: Hello from Harness via Modellix |
These examples are illustrative, not guaranteed verbatim; treat real output as authoritative. That’s it. The models named above bill at $0; paid models bill through your existing Modellix balance.
Beyond the plugin
dsh-modellix is one more way Modellix keeps your AI stack simple. The same key that powers your Harness conversations also works with our media generation API (image, video, audio) and the agent tools endpoint — one account, one balance, one bill.
- Try it today: modellix.ai
- Repo: github.com/Modellix/dsh-modellix
- Docs: docs.modellix.ai
Compliance and transparency
- Pricing transparency: unit costs are published; there are no hidden fees.
- Observability: every call is logged for debugging and performance tracking.
- Reliability: enterprise-grade infrastructure backed by over 15 years of developer-platform operations.
FAQ
Is the $0 billing permanent?
No. Select models are billed at $0 for now and may change. Always check the live catalog.
Do I need separate vendor accounts?
No. Modellix aggregates providers so you can operate with one key.
Will a gateway add latency?
We optimize routing and maintain transparent logs, with fallbacks and version-aware aliases to minimize integration overhead.
All capabilities and pricing referenced in this announcement are verified against Modellix documentation and live pricing as of August 28, 2026. Modellix is a unified AI gateway for image, video, audio and LLMs. Build faster with one key, transparent pricing, and enterprise-grade reliability. We have a commercial interest in the products described above.
