DAG state, deadlines, quality targets, budgets, reuse signals, and service policy.
Observe → decide → actuate → verify
RIDEControl
An open, agent-native control plane above LLM inference engines.
Agent workflows expose DAG state, tool waits, deadlines, quality bounds, reuse, and recovery intent. RIDE turns those signals and live telemetry into bounded fleet actions.
- 04
- decision domains
- 07
- incubating research lines
- 01
- evidence contract
System boundary
Engines execute. The control plane decides.
RIDE is not another inference engine. It is the policy, coordination, and evidence layer that turns workload intent and live telemetry into bounded actions on an engine fleet.
Observe, predict, route, admit, place, budget, recover, and verify.
vLLM-HUST, vLLM, SGLang, and compatible serving backends execute prefill, decode, cache, and communication paths.
The data plane remains independently measurable and replaceable.
Boundary rule
A RIDE mechanism must be expressible as a decision over observable state. Engine patches are allowed only as minimal reusable telemetry or actuation seams; kernels, KV transport, compilation internals, and hardware backends do not become control-plane work merely because they expose a flag.
Decision domains
Control is a closed loop, not a dashboard.
Each research line turns measurable state into a clear decision and a verifiable outcome.
Workflow & SLO
Schedule DAG stages, admit work, allocate capacity, and protect fairness along the critical path.
ADMISSION · DEADLINE · FAIRNESSRouting & placement
Choose a backend, cache holder, model replica, adapter, or continuation target from live fleet state.
SELECT · PLACE · RECOVERState & KV
Predict reuse, materialize or recompute state, select tiers, and account for transfer and survival horizons.
REUSE · MATERIALIZE · TIERQuality & economics
Trade quality, latency, cost, safety, and service risk without hiding failures behind aggregate throughput.
QUALITY · COST · GUARDRAILControl-plane portfolio
Independent questions. One control-plane architecture.
Each project retains its own owner, implementation, service, manifest, and results. RIDE provides the shared architectural family; repository links appear only after they are public.
Loading the portfolio…
Decision lifecycle
From workload intent to accountable execution.
RIDE connects observable workflow and system state to engine actions, then records outcomes for operations and evaluation.
Workload, topology, engine, model, queue, and resource signals.
Admission, routing, placement, budgets, state, quality, and recovery.
Apply decisions through versioned telemetry and control interfaces.
Correctness, latency, goodput, resource use, and fallback status.
Program leadership
Built as a coherent systems program.
Mao Yancan · yancanmao ↗
RIDE-Lab operations and Inference Control Plane program lead. Project-level academic ownership remains with each named owner.
Program coordination connects independent projects without replacing project-level accountability.Technical homes
One stack. Explicit ownership.
RIDE provides the control-plane architecture while each project retains the repository and organization that best match its technical ownership.