DSH AI Providers

Reusable, provider-neutral AI services for DeepSeek Harness.

DSH AI Providers is infrastructure for plugins that need AI capabilities beyond a single model backend. Users register a vendor connection and credential once, then reuse it for every capability that vendor supplies. Consumer plugins call stable services instead of implementing credentials, provider selection, concurrency control, and vendor APIs themselves.

It currently covers:

• language models;
• embeddings and reranking;
• text-to-speech and speech-to-text;
• image generation.

The project is primarily useful to plugin developers and to users of plugins built on it. Installing it alone does not add an assistant or start making AI requests: configure the providers you need, then install or build a plugin that consumes the services.

Why use it

• Configure a connection and DSH-managed credential once instead of repeating secrets in every feature or plugin.
• Let users choose or change providers without coupling consumer plugins to vendor-specific APIs.
• Share capability contracts, live configuration, and concurrency limits across plugins.
• Retain vendor-specific controls where a common interface would otherwise hide useful behavior.

The provider-neutral contracts are the foundation. A provider can be added behind those contracts without requiring each consumer plugin to add another vendor transport.

Documentation

• Documentation
• 简体中文文档
• Installation
• Configuration
• Using services from another plugin
• Core contract reference
• Creating a provider package

Packages

Package	Purpose
@dsh-ai-providers/core	Provider-neutral service contracts, errors, and shared types
@dsh-ai-providers/openai-compatible	Reusable OpenAI-compatible wire transports; not a plugin
@dsh-ai-providers/runtime	Provider registry, service selection, concurrency control, API routes, and settings UI
@dsh-ai-providers/openai	OpenAI-compatible language models, embeddings, TTS, file and live STT, and image generation
@dsh-ai-providers/localai	LocalAI language models, embeddings, reranking, TTS, file and live STT, and image generation
@dsh-ai-providers/anthropic	Language models through Anthropic's native Messages API
@dsh-ai-providers/jina-cohere	Reranking through Jina- and Cohere-compatible APIs
@dsh-ai-providers/all	Metadata-only DSH bundle containing the runtime and every provider maintained by this project

See the individual provider guides for configuration and supported features:

• OpenAI-compatible
• LocalAI
• Anthropic
• Jina and Cohere

Installation

Install the bundle into a DSH profile:

dsh plugin --profile <profile> add @dsh-ai-providers/all

The bundle installs the runtime and all provider packages maintained by this project. Restart the profile, then configure connections, routes, and active services under Settings > AI Providers. Individual packages may also be installed separately for a custom deployment.

To use these language-model routes in place of DSH's stock routes, disable @deepseek-ai/dsh-llm-pi-ai and the stock ui-settings-models settings page in that profile. Keep DSH's core llm service enabled. For DSH Voice Input, keep its speech-to-text registry enabled but select the API-backed provider instead of the built-in SenseVoice provider.

This is an independent project maintained by Firefox2100. It is not an official DeepSeek Harness project and is not maintained or endorsed by the DSH maintainers.

Development

Requires Node.js 22 or later and pnpm.

pnpm install
pnpm run typecheck
pnpm run test
pnpm run build
pnpm run release:check

See Contributing for branch and release policy.

License

GPL-3.0-only