feat: add LiteLLM as AI gateway provider

[RheagalFire]

Summary

• Adds LiteLLM as both an LLM and embedding provider, giving DeepSearcher users unified access to 100+ LLM providers (OpenAI, Anthropic, Google, Azure, Bedrock, Ollama, and more)
  through a single interface.
• LiteLLM is added as an optional dependency (deepsearcher[litellm]), keeping the default install lightweight.

Motivation

DeepSearcher already supports 18 LLM providers, each requiring a separate SDK and configuration. LiteLLM provides a unified gateway that routes to any provider using a single API,
reducing the need to maintain per-provider code. Users who already run a LiteLLM proxy get instant access to all their configured models.

Changes

• deepsearcher/llm/litellm_llm.py - New LiteLLM class implementing BaseLLM
• deepsearcher/embedding/litellm_embedding.py - New LiteLLMEmbedding class implementing BaseEmbedding
• deepsearcher/llm/__init__.py - Registered LiteLLM provider
• deepsearcher/embedding/__init__.py - Registered LiteLLMEmbedding provider
• pyproject.toml - Added litellm>=1.80.0,<1.87.0 as optional dependency
• deepsearcher/config.yaml - Added commented configuration examples
• tests/llm/test_litellm.py - 10 unit tests for LLM provider
• tests/embedding/test_litellm_embedding.py - 8 unit tests for embedding provider

Proof of work

Unit tests (18/18 passing):
tests/llm/test_litellm.py::TestLiteLLM::test_chat_multiple_messages PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_chat_omits_credentials_when_not_set PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_chat_single_message PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_chat_with_api_key PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_chat_with_error PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_drop_params_always_true PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_init_default PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_init_with_api_key_from_env PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_init_with_api_key_parameter PASSED
tests/llm/test_litellm.py::TestLiteLLM::test_init_with_custom_model PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_dimension_property PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_embed_documents PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_embed_query PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_embed_query_omits_credentials_when_not_set PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_embed_query_with_credentials PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_init_default PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_init_with_env_vars PASSED
tests/embedding/test_litellm_embedding.py::TestLiteLLMEmbedding::test_init_with_parameters PASSED

18 passed

Lint: ruff check and ruff format --check both pass clean.

Implementation notes

• drop_params=True is always set so provider-unsupported kwargs (e.g., seed, frequency_penalty on Anthropic) are silently dropped instead of causing errors.
• api_key and api_base are only forwarded when set. When omitted, LiteLLM reads provider-specific environment variables automatically (e.g., ANTHROPIC_API_KEY, OPENAI_API_KEY).
• litellm is lazy-imported inside method bodies so users who don't install it are unaffected.

Risk / Compatibility

• Additive only. Existing providers untouched.
• litellm is an optional extra - base install unaffected.

Example usage

# config.yaml
provide_settings:
  llm:                                                                                                                                                                                  
    provider: "LiteLLM"
    config:                                                                                                                                                                             
      model: "anthropic/claude-sonnet-4-20250514"
      # api_key: "sk-xxxx"  # or set ANTHROPIC_API_KEY env var                                                                                                                          

  embedding:                                                                                                                                                                            
    provider: "LiteLLMEmbedding"
    config:                                                                                                                                                                             
      model: "text-embedding-ada-002"
      dimension: 1536

  from deepsearcher.llm import LiteLLM
  from deepsearcher.embedding import LiteLLMEmbedding                                                                                                                                     
  
  # LLM - uses ANTHROPIC_API_KEY from environment                                                                                                                                         
  llm = LiteLLM(model="anthropic/claude-sonnet-4-20250514")
  response = llm.chat([{"role": "user", "content": "What is 2+2?"}])                                                                                                                      
  print(response.content)                                                                                                                                                                 
  
  # Embedding - uses OPENAI_API_KEY from environment                                                                                                                                      
  embedding = LiteLLMEmbedding(model="text-embedding-ada-002")
  vector = embedding.embed_query("Hello world")                                                                                                                                           
  print(f"Dimension: {len(vector)}")

[sre-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: RheagalFire
To complete the pull request process, please assign zc277584121 after the PR has been reviewed.
You can assign the PR to them by writing /assign @zc277584121 in a comment when ready.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[sre-ci-robot]

Welcome @RheagalFire! It looks like this is your first PR to zilliztech/deep-searcher 🎉

[RheagalFire]

/assign @zc277584121