Open source by Santander AI Lab. A tiny, vendor-neutral LLM client library — one interface for OpenAI, AWS Bedrock and Google Gemini (or bring your own AI backend).

Part of Santander AI Open Source — open source AI projects from Banco Santander (santander.com).

A tiny, vendor-neutral wrapper for any LLM backend. One small interface, pluggable providers. Write your application against LLMClient once and switch between OpenAI, AWS Bedrock, Google Gemini, a local server, or your own internal backend — without touching your code.

• Canonical interface + thin SDK adapters. One contract (LLMClient) with a small, modular adapter per official vendor SDK. The SDKs do the heavy lifting (retries, streaming, types); llm_bridge just normalises the calls.
• Zero required dependencies. The core (mock, callable) is pure standard library. Each vendor SDK is an optional extra you install on demand.
• Bring your own backend. Wrap any function with the callable provider — the library never needs to know which backend you use.
• One interface. chat(messages) and complete(prompt) returning a normalised LLMResponse (content + token usage + latency).

pip install llm-bridge                 # core only (no vendor SDKs)
pip install "llm-bridge[openai]"       # + OpenAI SDK
pip install "llm-bridge[aws]"          # + AWS Bedrock (boto3)
pip install "llm-bridge[google]"       # + Google Gemini (google-genai)
pip install "llm-bridge[all]"          # everything

from llm_bridge import create_llm

# Offline, no credentials — great for tests and demos.
llm = create_llm({"provider": "mock"})
print(llm.complete("Hello!").content)

# Switch provider by changing one dict — your code stays the same.
llm = create_llm({"provider": "openai", "model": "gpt-4o-mini"})       # needs [openai] + OPENAI_API_KEY
llm = create_llm({"provider": "bedrock", "model": "<bedrock-model-id>"}) # needs [aws] + AWS creds
llm = create_llm({"provider": "google", "model": "gemini-2.5-flash"})   # needs [google] + GOOGLE_API_KEY

resp = llm.chat([
    {"role": "system", "content": "You are concise."},
    {"role": "user", "content": "Name three primary colors."},
])
print(resp.content, resp.total_tokens)

The callable provider wraps any function — the recommended way to plug in a proprietary or internal backend:

from llm_bridge import create_llm

def my_backend(messages, temperature=0.7, max_tokens=1024, **kwargs):
    # call your own SDK / gateway / local model here, return the text
    return "the model output"

llm = create_llm({"provider": "callable", "callable": my_backend})
print(llm.complete("Hi").content)

Credentials are read from environment variables (OPENAI_API_KEY, GOOGLE_API_KEY/GEMINI_API_KEY, standard AWS credential chain). Never hardcode secrets.

The openai provider also targets any OpenAI-compatible endpoint (vLLM, Ollama, Azure OpenAI, or an internal gateway): pass a base_url (or set OPENAI_BASE_URL). For local servers without auth, set a dummy OPENAI_API_KEY.

class LLMClient:
    @property
    def model(self) -> str: ...
    @property
    def provider(self) -> str: ...
    def chat(self, messages, *, temperature=0.7, max_tokens=1024, **kwargs) -> LLMResponse: ...
    def complete(self, prompt, *, system=None, temperature=0.7, max_tokens=1024, **kwargs) -> LLMResponse: ...

@dataclass
class LLMResponse:
    content: str
    model: str
    prompt_tokens: int
    completion_tokens: int
    finish_reason: str
    latency_ms: float
    raw: Any            # the provider's raw response
    # .total_tokens

Implement LLMClient, expose build(config) -> LLMClient, and register it in llm_bridge.registry. See CONTRIBUTING.md.

See examples/: mock_example.py, callable_example.py, openai_example.py, bedrock_example.py, google_example.py.

• Python 3.9+
• No required runtime dependencies for the core (mock, callable).
• Optional vendor SDKs are installed on demand via extras ([openai], [aws], [google], [all]).

Contributions are welcome! Please read CONTRIBUTING.md and our CODE_OF_CONDUCT.md. External contributors will be asked to sign the CLA (handled automatically by the CLA Assistant bot).

Please report vulnerabilities privately — see SECURITY.md. Do not open a public issue for security reports.

If you use this software, please cite it using the metadata in CITATION.cff.

Apache License 2.0 — see LICENSE and NOTICE.