bitsky-tech / bridgic-llms

Bridgic LLMs

Model-neutral LLM integration with protocol-driven capability declaration.

Dependencies

Install only the LLM provider package you need. python-dotenv is required for loading .env configuration.

Installation: Run the install script to set up all dependencies:

bash "skills/bridgic-llms/scripts/install-deps.sh" "$PWD" [PROVIDER]

Supported providers: openai (default), openai-like, vllm. The script checks uv availability, initializes a uv project if needed, installs any missing packages via uv add, and runs uv sync to finalize the environment. When it exits successfully the project is fully initialized and ready to use — no manual uv add / uv sync follow-up is required.

Quick Start

import os
from dotenv import load_dotenv
from bridgic.llms.openai import OpenAILlm, OpenAIConfiguration

load_dotenv()

llm = OpenAILlm(
    api_key=os.environ.get("LLM_API_KEY"),
    api_base=os.environ.get("LLM_API_BASE"),
    configuration=OpenAIConfiguration(
        model=os.environ.get("LLM_MODEL", "gpt-4o"),
        temperature=0.0,
        max_tokens=16384,
    ),
    timeout=180.0,
)

Provider Selection Guide

Common pitfall: Do NOT use OpenAILikeLlm when you need structured output or tool selection — it does not implement those protocols. Use OpenAILlm instead.

Basic Interfaces

All providers implement BaseLlm:

from bridgic.core.model.types import Message, Role

messages = [
    Message.from_text("You are a helpful assistant.", role=Role.SYSTEM),
    Message.from_text("Hello!", role=Role.USER),
]

# Chat — complete response
response = llm.chat(messages=messages, model="gpt-4o", temperature=0.7)
print(response.message.content)

# Stream — real-time chunks
for chunk in llm.stream(messages=messages, model="gpt-4o"):
    print(chunk.delta, end="", flush=True)

Advanced Protocols

See [references/llm-integration.md](references/llm-integration.md) for:

• StructuredOutput — generate Pydantic model instances or JSON schema conformant output
• ToolSelection — function/tool calling with Tool definitions
• Full code examples for all providers

Reference Files