AccuralAI Core
The orchestration nucleus for the AccuralAI local LLM ecosystem.
Overview
accuralai-core provides the async stage runner that orchestrates the entire LLM pipeline. Requests flow through:
Canonicalization → Cache lookup → Router selection → Backend invocation → Validation → Optional post-processing
Stage metadata is recorded on an ExecutionContext for tracing, cancellation, and event publishing.
Key Components
Pipeline Architecture
The core pipeline (accuralai_core.core.pipeline) defines the async stage runner that processes requests through canonical stages.
Context Management
ExecutionContext (accuralai_core.core.context) provides tracing, cancellation, and event publishing capabilities.
Plugin System
Plugin groups follow the accuralai_core.<stage> naming convention and are auto-discovered via importlib entry points.
API Reference
Entry Points
The core package registers several entry points:
accuralai_core.backends: Backend implementationsaccuralai_core.validators: Validation pluginsaccuralai_core.caches: Caching implementations
Configuration
See the configuration example for setup details.
CLI Usage
# Generate text using the CLI
accuralai generate "Hello, world!"
# With metadata
accuralai generate "Hello, world!" --metadata user=developer --metadata session=test
# With parameters
accuralai generate "Hello, world!" --param temperature=0.7 --param max_tokens=100
Development
# Install in development mode
pip install -e packages/accuralai-core[dev]
# Run tests
pytest packages/accuralai-core/tests/
# Run linting
ruff check packages/accuralai-core/