DSH AI Providers
DSH AI Providers gives DeepSeek Harness plugins a shared way to use AI services. Configure a provider connection and a DSH-managed credential once, then reuse them across supported capabilities and consumer plugins. Plugin developers work against provider-neutral service contracts instead of rebuilding credential forms, provider selection, concurrency control, and vendor transports.
The current services are:
- embeddings and reranking;
- typed, probability-backed decisions;
- text-to-speech and speech-to-text;
- image generation;
- language-model routes.
Install the runtime and the provider plugins you need. Configure connections in DSH, select a provider for each capability, then consume the selected service from the Cordis context.
This is infrastructure for other plugins. Installing it does not add an assistant or send an AI request by itself. Requests begin only when a consumer plugin calls one of the services. Because the consumer depends on the shared contract, users can change providers without requiring that plugin to understand every vendor API.
Provider-specific controls stay with their provider package. The shared contracts remain focused on behavior that consumers can use across implementations.
Included packages
| Package | Purpose |
|---|---|
@dsh-ai-providers/core | Provider-neutral contracts, errors, and shared types |
@dsh-ai-providers/openai-compatible | Shared OpenAI-compatible wire transports; not a plugin |
@dsh-ai-providers/jina-cohere | Shared Jina- and Cohere-style rerank and embedding transports; not a plugin |
@dsh-ai-providers/systemone | Shared System One decision transport; not a plugin |
@dsh-ai-providers/runtime | Provider registry, selected services, settings UI, API routes, and concurrency control |
@dsh-ai-providers/openai | OpenAI-compatible embeddings, LLM, TTS, STT, and image generation |
@dsh-ai-providers/localai | LocalAI embeddings, reranking, decisions, LLM, TTS, STT, and image generation |
@dsh-ai-providers/anthropic | Anthropic-native language models |
@dsh-ai-providers/ollama | Ollama-native language models and embeddings |
@dsh-ai-providers/cohere | Cohere-native language models, embeddings, and reranking |
@dsh-ai-providers/dashscope | DashScope-native (Qwen) language models, embeddings, reranking, speech, and images |
@dsh-ai-providers/jina | Jina reranking and embeddings |
@dsh-ai-providers/typesafe | TypeSafe's Jev decisions |
@dsh-ai-providers/all | Metadata-only bundle containing the runtime and all providers maintained by this project |
Capability matrix
This table shows what each project-maintained vendor package registers. It describes the package, not every server behind it: an OpenAI-compatible or LocalAI deployment may implement only some of the listed endpoints.
| Provider | LLM | Embeddings | Reranking | Decisions | TTS | STT | Images |
|---|---|---|---|---|---|---|---|
| OpenAI-compatible | Yes | Yes | No | Yes | Yes | File and live | Yes |
| LocalAI | Yes | Yes | Yes | Yes | Yes | File and live | Yes |
| Anthropic | Yes | No | No | No | No | No | No |
| Ollama | Yes | Yes | No | No | No | No | No |
| Cohere | Yes | Yes | Yes | No | No | No | No |
| DashScope | Yes | Yes | Yes | No | Yes | File | Yes |
| Jina | No | Yes | Yes | No | No | No | No |
| TypeSafe | No | No | No | Yes | No | No | No |
Where to go next
- If you are building a plugin that consumes AI services, start with Using services and keep the core contract reference nearby.
- If you are adding a vendor implementation, read Creating a provider.
- If you are configuring a DSH profile, use Installation, then Configuration.
Providers maintained by this project
These packages are maintained by Firefox2100 as part of this independent project. They are not official DeepSeek Harness packages and are not maintained or endorsed by the DSH maintainers.
This project currently maintains eight vendor packages:
@dsh-ai-providers/openai: OpenAI-compatible language models, embeddings, decisions, TTS, file and live STT, and image generation;@dsh-ai-providers/localai: dedicated LocalAI language models, embeddings, reranking, decisions, speech, and image generation;@dsh-ai-providers/anthropic: language models through Anthropic's native Messages API;@dsh-ai-providers/ollama: language models and embeddings through Ollama's native API, with its own request controls, discovery, and model load and unload;@dsh-ai-providers/cohere: language models, embeddings, and reranking through Cohere's native v2 API;@dsh-ai-providers/dashscope: language models, embeddings, reranking, speech synthesis and recognition, and image generation through Alibaba Cloud DashScope's native API (Qwen);@dsh-ai-providers/jina: Jina reranking and embeddings;@dsh-ai-providers/typesafe: decisions from TypeSafe's hosted Jev model.
Third-party provider plugins can implement the same contracts and appear in the same settings UI.