NEXOS CORE COMPOSE RUNTIME
Composable Primitives & Universal Model Adapters
The high-performance composable abstraction layer for modern AI engineering. Unified chat interfaces, structured output parsers, and zero-tax runtime adapters across every major frontier and local LLM.
Universal Model & Hardware Agnosticism
100% Interoperability| Provider | Supported Architectures | Adapter Overhead | Status |
|---|---|---|---|
| OpenAI | GPT-4o, o1, o3-mini | 0.12ms adapter tax | NATIVE |
| Anthropic | Claude 3.5 Sonnet, 3.7 Opus | 0.10ms adapter tax | NATIVE |
| Gemini 2.0 Pro, 2.0 Flash | 0.11ms adapter tax | NATIVE | |
| DeepSeek | DeepSeek R1, DeepSeek V3 | 0.12ms adapter tax | NATIVE |
| Local / Self-Hosted | vLLM, Ollama, SGLang, Llama 3 | 0.04ms adapter tax | NATIVE |
composable_pipeline.py
LCEL-Style Composable Syntaxfrom nexos.core import PromptTemplate, ChatAdapter, JsonOutputParser
# 1. Define model-agnostic prompt template
prompt = PromptTemplate.from_template(
"Extract financial risk factors for company: {company_name}"
)
# 2. Bind unified model adapter (seamlessly swap Claude, Gemini, or vLLM)
model = ChatAdapter(provider="anthropic", model="claude-3-7-opus")
parser = JsonOutputParser()
# 3. Pipe chain composition
chain = prompt | model | parser
# Execute with sub-millisecond serialization
output = chain.invoke({"company_name": "NEXOS Technologies"})
print("Parsed Output:", output)