A declarative, composable framework for building transparent LLM-powered systems through dataflow abstractions.
Option 1: HUST Campus Network Access 🎓
Our team maintains a live deployment accessible within HUST campus network:
🌐 Contact team for access URL
Requirements: HUST campus network or VPN connection
Experience SAGE's visual pipeline editor and AI-powered chat assistant with RAG capabilities!
Option 2: Local Installation
git clone https://github.com/intellistream/SAGE.git
cd SAGE
./quickstart.sh --standard --yes
sage studio start
# Visit http://localhost:4200
SAGE is a high-performance streaming framework for building AI-powered data processing pipelines. Transform complex LLM reasoning workflows into transparent, scalable, and maintainable systems through declarative dataflow abstractions.
See docs-public/docs_src/dev-notes/cross-layer/team-management.md for the current team
coordination entrypoint (team management + incubation policy).
Production-Ready: Built for enterprise-scale applications with distributed processing, fault tolerance, and comprehensive monitoring out of the box.
Developer Experience: Write complex AI pipelines in just a few lines of code with intuitive declarative APIs that eliminate boilerplate.
Performance: Optimized for high-throughput streaming workloads with intelligent memory management and parallel execution capabilities.
Transparency: Built-in observability and debugging tools provide complete visibility into execution paths and performance characteristics.
Flexible Deployment: Full support for CPU-only compute nodes alongside GPU nodes, with intelligent resource-aware scheduling for hybrid clusters.
Transform rigid LLM applications into flexible, observable workflows. Traditional imperative approaches create brittle systems:
# Traditional approach - rigid and hard to modify
def traditional_rag(query):
docs = retriever.retrieve(query)
if len(docs) < 3:
docs = fallback_retriever.retrieve(query)
prompt = build_prompt(query, docs)
response = llm.generate(prompt)
return responseSAGE transforms this into a declarative, composable workflow:
from sage.kernel.api.local_environment import LocalEnvironment
from sage.libs.io.source import FileSource
from sage.middleware.operators.rag import DenseRetriever, QAPromptor, OpenAIGenerator
from sage.libs.io.sink import TerminalSink
# Create execution environment
env = LocalEnvironment("rag_pipeline")
# Build declarative pipeline
(
env.from_source(FileSource, {"file_path": "questions.txt"})
.map(DenseRetriever, {"model": "sentence-transformers/all-MiniLM-L6-v2"})
.map(QAPromptor, {"template": "Answer based on context: {context}\nQ: {query}\nA:"})
.map(OpenAIGenerator, {"model": "gpt-3.5-turbo"})
.sink(TerminalSink)
)
# Execute pipeline
env.submit()Try it yourself:
git clone https://github.com/intellistream/SAGE.git && cd SAGE
git checkout main-dev
./quickstart.sh --dev --yes
python examples/tutorials/hello_world.pyFor CPU-only deployment:
# Start JobManager for distributed task execution
sage jobmanager start
# Run CPU node demo (no GPU required)
python examples/tutorials/L3-kernel/cpu_node_demo.pySAGE is built on a layered modular architecture with 11 independent packages organized across 6 layers:
L6: sage-studio, sage-cli, sage-tools # User Interfaces & Dev Tools
L5: sage-apps # Applications
L4: sage-middleware, sage-gateway # Domain Operators & API Gateway
L3: sage-kernel, sage-libs # Core Engine & Algorithm Library
L2: sage-platform # Platform Services (Queue, Storage)
L1: sage-common # Foundation & Utilities
Key Architectural Principles:
- Unidirectional Dependencies: Clean layer-to-layer dependencies (no upward dependencies)
- Separation of Concerns: Each package has a clear, focused responsibility
- Pluggable Components: Modular design allows easy component replacement
- Production Ready: Built-in fault tolerance, monitoring, and distributed execution
📖 Complete Architecture Guide - Detailed package descriptions, dependency rules, and design principles
11 Independent Packages, each with clear responsibilities:
- sage-common (L1): Foundation utilities, configuration, logging
- sage-platform (L2): Platform services - queue, storage abstractions
- sage-kernel (L3): Distributed execution engine and runtime
- sage-libs (L3): Algorithm library, RAG tools, Agent framework
- sage-middleware (L4): Domain operators and middleware components
- sage-gateway (L4): API gateway and service mesh
- sage-apps (L5): Pre-built applications (video, medical diagnosis)
- sage-studio (L6): Web-based visualization interface
- sage-cli (L6): Unified command-line interface
- sage-tools (L6): Development tools and testing framework
Note: sage-benchmark has been separated into an independent repository: https://github.com/intellistream/sage-benchmark
- Distributed Execution with automatic load balancing
- Fault Tolerance and error recovery
- Observability with metrics and monitoring
- Extensible Integration for databases, queues, and AI services
Quickstart (Recommended)
git clone https://github.com/intellistream/SAGE.git && cd SAGE
./quickstart.sh --dev --yes # Interactive mode: ./quickstart.sh⚡ Auto-Acceleration: Network optimization is now enabled by default:
- 🌐 Auto-detects network location (China mainland → mirror sources)
- 🚀 Parallel downloads (8 threads) + pre-compiled packages
- ⏱️ 3-5x faster installation: 12-18 min (vs 35-45 min)
- 🔧 Disable:
./quickstart.sh --no-mirror --dev --yes
PyPI Install
pip install isage[standard] # Recommended
pip install isage[core] # Minimal runtime
pip install isage[full] # Full features + Web UI
pip install isage[dev] # Development toolsVerification & Troubleshooting
sage doctor # Check installation
./quickstart.sh --doctor # Diagnose issues📖 Detailed guides: Installation Guide | Troubleshooting | Validation | Optimization Tips
cp .env.template .env # Copy template
# Edit .env and add your API keys (OPENAI_API_KEY, HF_TOKEN, etc.)📖 API key setup: See .env.template for all available options
RAG Applications: Build production-ready retrieval-augmented generation systems with multi-modal support and advanced reasoning capabilities.
Real-Time Analytics: Process streaming data with AI-powered insights, anomaly detection, and automated decision making.
Data Pipeline Orchestration: Coordinate complex ETL workflows that seamlessly integrate AI components with traditional data processing.
Multi-Modal Processing: Handle text, images, audio, and structured data in unified pipelines with consistent APIs. 🆕 Advanced multimodal fusion enables intelligent combination of different data modalities for enhanced AI understanding and generation.
Distributed AI Inference: Scale AI model serving across multiple nodes with automatic load balancing and fault tolerance.
- Documentation: https://intellistream.github.io/SAGE-Pub/
- Examples: examples/ - Tutorials, RAG, services, benchmarks
- Quick Reference: docs/QUICK_REFERENCE.md
- Architecture: docs-public/docs_src/dev-notes/package-architecture.md
We welcome contributions! See CONTRIBUTING.md for guidelines.
git checkout -b feature/my-feature
./quickstart.sh --dev --yes
# Make changes, add tests
sage-dev quality && sage-dev test
git commit -m "feat(kernel): add new feature"
git push -u origin feature/my-featureResources: Quick Reference | GitHub Issues | Discussions
make help # View all commands
sage-dev quality # Format & lint
sage-dev test # Run tests
make docs # Build documentation📖 Complete reference: docs/dev-notes/DEV_COMMANDS.md
💬 Join SAGE Community - WeChat, QQ, Slack, GitHub Discussions
SAGE is licensed under the MIT License.
![@github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=64&v=4){kind=small}
