wenzetan/dsh-llm-newapi
40Last commit Aug 15, 2026
dsh-llm-newapi DSH plugin
dsh-llm-newapi integrates the NewAPI gateway into DeepSeek Harness (DSH) as an LLM provider route. It provides a dedicated settings page in the DSH web panel for configuring API key, base URL, and fetching available models. The plugin automatically filters out non-chat models (e.g., embedding, rerank) and supports hot-reload via settings.yaml.
How to install the dsh-llm-newapi DSH plugin
dsh plugin --profile web add dsh-llm-newapiCopying does not run this command. Review the repository and version before installing the dsh-llm-newapi DSH plugin.
dsh-llm-newapi DSH plugin data source
dsh-llm-newapi DSH plugin snapshot date: Aug 16, 2026
discovered
What the dsh-llm-newapi DSH plugin can do
- Implements LlmAdapter seam for NewAPI (OpenAI-compatible chat completions and model listing).
- Web settings page with 'Fetch model info' to populate model catalog, API key input, and baseURL configuration.
- Automatic filtering of embedding/rerank/ranker models during discovery, configurable via modelExcludePatterns.
- Parameter matching engine with built-in vendor hints and configurable providerHints for official model entries.
- Supports stable and prerelease installation channels (npm and GitHub) with manual version pinning.
- Hot-reloadable configuration via cordis.yml and settings.yaml; credentials stored in DSH credentials service.
Where the dsh-llm-newapi DSH plugin fits
- Connect DSH to a self-hosted NewAPI gateway for LLM completions.
- Use third-party APIs that are compatible with the OpenAI API format (e.g., custom proxies, aggregated services).
- Test and compare different LLM models through a single gateway with unified configuration.
- Manage multiple LLM provider routes in DSH, supplementing built-in providers with a custom one.
Who the dsh-llm-newapi DSH plugin is for
- DSH users who need to add a custom LLM provider beyond the official ones.
- Developers or administrators managing DSH instances that require integration with a private or third-party OpenAI-compatible API.
dsh-llm-newapi DSH plugin limitations
- Requires a running NewAPI-compatible gateway (OpenAI API format with /v1 prefix).
- API key must be set via the web settings page; environment variable support is not provided due to credential service design.
- Plugin is still in early development (v0.8.x); breaking changes may occur between versions.
- Model discovery and configuration are manual; the model list must be refreshed when the upstream changes.
dsh-llm-newapi DSH plugin: from the repository README
Quoted from the wenzetan/dsh-llm-newapi README, the upstream source of the dsh-llm-newapi DSH plugin. Copyright remains with the original authors.
**English** | [中文](README.zh-CN.md) An LLM provider plugin that adds **NewAPI** to [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) (dsh). **Zero modifications to dsh itself.** - Provider route id: `newapi` - Display name: `NewAPI` - Shape: LLM Provider plugin — implements the `LlmAdapter` seam from `@deepseek-ai/dsh-llm`; NewAPI is an OpenAI-compatible gateway (`POST {baseURL}/chat/completions`, `GET {baseURL}/models`, baseURL includes `/v1`) - Dual-face structure: host side (adapter + model discovery) + browser side (a "NewAPI" settings page in the dsh web settings panel, including "Fetch model info") Design decisions and trade-off analysis live in [DESIGN.md](DESIGN.md); reference implementation: `deepseek-harness/packages/llm/llm-deepseek`. ## Installation (dsh ≥ 0.1.0-rc) ### Release channels and version selection The project uses a **dual release channel**. Install from the **npm registry** (default) or a **versioned tag**; never from the `main` branch HEAD: | Channel | Version shape | Install reference | When to use | |---|---|---|---| | **Stable (recommended)** | `vX.Y.Z` | `dsh plugin --profile web add dsh-llm-newapi` (npm `latest`) or `github:wen
Read the full READMERepository license: MIT
dsh-llm-newapi DSH plugin questions
What is dsh-llm-newapi?
dsh-llm-newapi is a plugin for DeepSeek Harness (DSH) that adds support for NewAPI, an OpenAI-compatible gateway. It allows you to use any backend that implements the OpenAI chat completions API (e.g., local proxies, aggregated services) as a provider in DSH. The plugin includes a settings page in the DSH web interface for configuration and model discovery.
How do I install dsh-llm-newapi?
Use the DSH CLI command: `dsh plugin --profile web add dsh-llm-newapi`. This installs the latest stable version from npm. For prerelease versions, append `@next`. After installation, you must register the bundle in your profile's package.json by adding `"dsh-llm-newapi"` to the `dsh.profile.bundles` array, then restart DSH web. Refer to the README for alternative methods like GitHub shorthand or tarball.
How do I configure the NewAPI plugin?
After installation, a 'NewAPI' settings page will appear in the DSH web panel. Fill in the API key and the gateway base URL (including `/v1`, e.g. `http://gw.local:3000/v1`). Click 'Fetch model info' to pull the model list, then select the chat models you want to use. You can also configure via `cordis.yml` or `settings.yaml` for advanced options like `modelExcludePatterns` and `providerHints`. The configuration hot-reloads without restart.
Can I use any OpenAI-compatible API with this plugin?
Yes, as long as the API follows the OpenAI chat completions format (`POST /v1/chat/completions`) and exposes a `GET /v1/models` endpoint. The plugin is designed to work with any gateway that implements these endpoints. However, note that the API key is stored in DSH's credentials service and must be entered via the web settings page, not via environment variables.
How do I update dsh-llm-newapi to the latest version?
Run `dsh plugin --profile web update dsh-llm-newapi` to update to the latest stable version. If you want to switch to a specific version, use the `add` command with a version tag, e.g., `dsh plugin --profile web add dsh-llm-newapi@1.0.0`. After updating, you may need to restart DSH web. The plugin's built-in release channels are described in the README: stable (latest) and prerelease (next).