SAGE — Streaming-Augmented Generative Execution — is the shared flagship product and technical vision of the IntelliStream research ecosystem. It applies streaming-computing principles to LLM inference and agent execution.
SAGE is a product boundary, not a fifth research organization. Its public surface integrates work across agent systems, evolving-data infrastructure, and model execution while keeping each implementation boundary explicit.
The SAGE core series is hosted under the RIDE-Lab GitHub namespace because RIDE
Lab is its principal engineering steward. RIDE Lab maintains the core framework,
agent-facing product surfaces, reviews, releases, and product documentation.
Namespace placement is an operational home, not a claim of exclusive academic or product ownership. SAGE remains a shared ecosystem product.
- IntelliStream incubates early cross-layer research until its public abstraction, maintainers, evaluation path, and durable home are clear.
- RIDE Lab owns agent-native research systems and stewards SAGE core, orchestration, RAG, evaluation, studio, and application-facing integration.
- DataSys owns framework-neutral stream, graph, vector/index, evolving-data, query, lifecycle, and data-system benchmark implementations.
- vLLM-HUST owns model execution, KV-cache mechanisms, decode scheduling, compilation, kernels, hardware backends, and inference validation.
SAGE consumes DataSys capabilities and calls vLLM-HUST through documented interfaces. Shared product language never transfers implementation ownership across these boundaries.
- Decisions remain with the maintainers of the repository that owns the affected implementation.
- Cross-boundary changes should define an interface or integration contract and be reviewed by maintainers on both sides.
- SAGE product documentation should distinguish shared product vision, core stewardship, and dependency ownership.
- Moving a repository between organizations requires a durable ownership change; product integration alone is not sufficient.
Applications such as Sage Mate are built with SAGE and use vLLM-HUST or another explicitly configured inference service for model execution. Application placement does not alter SAGE's shared-product status.