Other partners

Quick reference for the 13 partner packages besides OpenAI and Anthropic. Each follows the same structural conventions described in partners/index: a chat_models.py (or directory), optional embeddings.py/llms.py/tools/, a data/ directory of model profiles, and a test suite that inherits from langchain-tests.

Ollama

libs/partners/ollama/. PyPI: langchain-ollama. Local model support via the Ollama runtime.

• Exports: ChatOllama, OllamaLLM, OllamaEmbeddings.
• Talks to a local Ollama server over HTTP. No API key.
• Supports tool calling for models that expose it (e.g. recent Llama and Mistral builds).
• Tests run against a real Ollama instance in CI when secrets allow; otherwise, mocks via responses.

Groq

libs/partners/groq/. PyPI: langchain-groq. The Groq inference API.

• Exports: ChatGroq.
• Notable for sub-100ms time-to-first-token on supported models.
• Inherits the OpenAI-compatible Chat Completions surface, so its implementation is small and mostly delegates to groq>=0.x SDK.

Mistral AI

libs/partners/mistralai/. PyPI: langchain-mistralai. The Mistral cloud API.

• Exports: ChatMistralAI, MistralAIEmbeddings.
• Supports Mistral's tool calling, function calling, and the mistral-large and mistral-small model families.
• Streaming with usage metadata.

Hugging Face

libs/partners/huggingface/. PyPI: langchain-huggingface. Bridges LangChain to the Hugging Face ecosystem.

• Exports: ChatHuggingFace, HuggingFaceEmbeddings, HuggingFacePipeline, HuggingFaceEndpoint.
• Supports both local inference (transformers pipelines, text-generation-inference) and hosted Inference API.
• The _BUILTIN_PROVIDERS registry uses a custom constructor (from_model_id(model_id=...)) because Hugging Face models are identified by repo IDs, not provider model names.

DeepSeek

libs/partners/deepseek/. PyPI: langchain-deepseek. The DeepSeek API.

• Exports: ChatDeepSeek.
• Notable for deepseek-reasoner, which produces reasoning tokens mapped to LangChain's ReasoningContentBlock.

Fireworks

libs/partners/fireworks/. PyPI: langchain-fireworks. Fireworks AI's inference API.

• Exports: ChatFireworks, FireworksEmbeddings, Fireworks (LLM).
• Supports tool calling, JSON mode, and multimodal content for compatible models.

OpenRouter

libs/partners/openrouter/. PyPI: langchain-openrouter. A meta-provider that routes to many backends through one API.

• Exports: ChatOpenRouter.
• Mostly an OpenAI-compatible Chat Completions wrapper with provider-routing extensions in headers.

Perplexity

libs/partners/perplexity/. PyPI: langchain-perplexity. The Perplexity AI API.

• Exports: ChatPerplexity.
• Adds web search–grounded responses with citation content blocks.

xAI

libs/partners/xai/. PyPI: langchain-xai. The xAI / Grok API.

• Exports: ChatXAI.
• OpenAI-compatible surface; the implementation is correspondingly thin.

Exa

libs/partners/exa/. PyPI: langchain-exa. Exa's neural search API.

• Exports: ExaSearchRetriever, ExaSearchResults, ExaFindSimilar.
• This partner is a retriever and tool rather than a chat model — it gives agents the ability to search the web through Exa.

Nomic

libs/partners/nomic/. PyPI: langchain-nomic. Nomic's embedding API.

• Exports: NomicEmbeddings.
• Embedding-only — no chat model.

Chroma

libs/partners/chroma/. PyPI: langchain-chroma. The Chroma vector store.

• Exports: Chroma, the VectorStore implementation backed by a local or remote Chroma database.
• Used in many RAG examples; one of the most-installed partner packages.

Qdrant

libs/partners/qdrant/. PyPI: langchain-qdrant. Qdrant vector database.

• Exports: QdrantVectorStore, FastEmbedEmbeddings (a CPU embedding model that pairs well with Qdrant).
• Supports hybrid search, payload filtering, sparse vectors, and the FastEmbed local model.

Common conventions

Every package above:

• Lives at libs/partners/<name>/langchain_<name>/.
• Re-exports its public classes from __init__.py.
• Implements langchain-core interfaces (BaseChatModel, Embeddings, VectorStore, …).
• Ships profile data in data/ refreshed by langchain-profiles.
• Inherits standard tests from langchain-tests.
• Has a Makefile with test, integration_test, lint, format targets.
• Releases independently via .github/workflows/_release.yml.

Related

• partners/index
• partners/openai, partners/anthropic — full pages on the two largest partners
• packages/standard-tests
• packages/model-profiles