Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture, allowing developers to extend model capabilities by adding external LLM providers. The dsh-opencode-zen plugin is designed to address the need for free models with tool-calling and reasoning capabilities when there is no budget or when configuring complex API keys is undesirable. It directly integrates OpenCode Zen’s free models into DSH’s model selector.

Plugin Information

  • Name: dsh-opencode-zen
  • Owner: keman-ai
  • Category: Model Inference
  • License: MIT
  • Core value: Zero configuration, no API key required, real-time model catalog retrieval from upstream

Core Features

  • Tool-calling and reasoning support: All 7 integrated models support tool calling and reasoning content, enabling a full agent loop rather than chat-only functionality.
  • Zero-configuration operation: Works out of the box with anonymous calls by default, without requiring account registration or balance top-ups.
  • Dynamic catalog: The model list is not hardcoded but is fetched dynamically from upstream at runtime to ensure model availability.

Installation and Enablement

The plugin is not published to npm and must be installed from GitHub.

  1. Run the following command to install the plugin (make sure to use the -w flag, otherwise an error will occur):
    dsh plugin --profile web add -w github:keman-ai/dsh-opencode-zen

*   The `--profile` parameter should be selected based on the current DSH usage scenario (`web` or `headless`).
*   After installation, restart DSH.
  1. After restarting, the opencode-zen model group will appear in the model selector.

Model List

The plugin provides the following 7 free models by default (the specific list may change as upstream updates are released):

  • nemotron-3-ultra-free: 1M context
  • nemotron-3.5-lightning-free: 256K context, 256K output
  • laguna-s-2.1-free: 256K context
  • deepseek-v4-flash-free: 200K context, 128K output
  • big-pickle: Zen’s own anonymous evaluation model
  • mimo-v2.5-free: 200K context
  • hy3-free: 190K context

Configuration

Plugin configuration is fully optional. Leave it empty to use the default behavior.

If you need to use a private quota or custom parameters, you can edit the configuration file:

plugins:
  dsh-opencode-zen:
    apiKeyEnv: OPENCODE_API_KEY        # Credential reference (environment variable name)
    baseURL: https://opencode.ai/zen/v1
    catalogUrl: https://models.dev/api.json
    catalogTtlMs: 3600000              # Catalog cache duration, default 1 hour
    catalogTimeoutMs: 8000             # Catalog request timeout
    maxTokens: 32000                   # Output length limit, the model's own limit will take effect
    defaultContextWindow: 128000       # Default context window when no catalog entry exists

If you are using a private quota, set the environment variable or manage the key through DSH’s credential service:

export OPENCODE_API_KEY=<your key>

Use Cases and Limitations

Use Cases

  • Developers who need to test tool-calling and complex reasoning capabilities.
  • Users who want to run agents with minimal cost and minimal configuration.

Considerations

  • Shared quota limits: The free quota is a shared resource, and anonymous calls can easily trigger rate limits. It is recommended to switch to key-based mode before the quota is exhausted.
  • Reasoning content is not round-tripped: Reasoning is displayed only in the conversation and is not sent back to the model as history (limited by the OpenAI-compatible interface).
  • Text only: All models support text input only, and image inputs are replaced with a single-line placeholder.
  • Paid models excluded: The plugin includes only free models. Paid models from OpenCode Zen (such as Claude, GPT, etc.) are not included.

Development and Maintenance

  • The plugin is not published to npm, and the source code is available on GitHub.
  • Build artifacts (lib/) have been committed to the repository, so it can be installed directly from source without a local build toolchain.
  • Development environment setup:
    git clone https://github.com/keman-ai/dsh-opencode-zen
    cd dsh-opencode-zen && pnpm install && pnpm build
    dsh plugin --profile web add <绝对路径>

Summary

dsh-opencode-zen provides DeepSeek Harness users with a direct path to OpenCode Zen’s free models. By dynamically retrieving the model list and supporting zero-configuration startup, it lowers the barrier to using advanced LLM capabilities, making it suitable for rapid prototyping or cost-sensitive agent tasks.