AEF — AIVA Execution Fabric
Physics layer of AIVA: particle-based execution engine
Part of the AIVA archetype. ← Back to AIVA
Role
AEF (AIVA Execution Fabric) is the Physics layer of the triadic AIVA system. It receives compiled DAGs from AQL, turns them into particle instructions, and runs them: scheduling, resource allocation, and telemetry are all within AEF. AIOS decides what to do; AQL proves the structure is correct; AEF executes that structure. Particles are stateless (A2) and deterministic (A1), so execution is reproducible and auditable. The physics layer exists because running is a different job from deciding and proving: splitting execution from structure gives scale (particles run in parallel, across nodes, without the brain or chemistry in the loop), resource safety (AEF enforces concurrency and time budgets; admission control rejects overload per A9), and observability (every run produces canonical execution state and metrics, fed back to AIOS via the ANS for monitoring and fail-close). AEF is bounded, observable execution that the rest of the pipeline can trust and audit.
Architecture
AEF organizes work into particle types, each with a well-defined role and contract. Quarks perform core computation (e.g. LOAD_QUARK, EXEC_QUARK, STORE_QUARK). Leptons handle I/O (READ_LEPTON, WRITE_LEPTON, STREAM_LEPTON). Bosons handle communication and messaging (EMIT_BOSON, RECV_BOSON, BROADCAST_BOSON). Gluons handle binding and synchronization (BIND_GLUON, RELEASE_GLUON, BARRIER_GLUON). Neutrinos handle silent monitoring and tracing (TRACE_NEUTRINO, MONITOR_NEUTRINO, LOG_NEUTRINO). The scheduler respects DAG dependencies and resource budgets so that particles run in the correct order without overloading the system. Execution is quantum-inspired on classical hardware—superposition and entanglement are simulated where useful—with energy-based resource management and built-in telemetry so that every run is observable and, where required, reproducible.
Key features
AEF provides quantum-like computation on classical hardware (superposition and entanglement simulation where it aids parallelism or reasoning), energy-based resource management so that CPU, memory, and time are bounded (A9), and built-in telemetry and observability so that execution state and metrics are canonical and feed back into AIOS. A/B testing and optimization can be run over particle execution so that tuning is data-driven and consistent with the same mathematical guarantees that govern the rest of the pipeline.
Integration
AQL feeds AEF with compiled DAGs. AEF turns those DAGs into particle instructions, schedules them, and runs them. No structure verification happens in AEF; AQL guarantees correctness before execution begins.
AIOS receives execution telemetry from AEF via the SNS and ANS. Telemetry (latency, errors, resource use) is used for monitoring, fail-close (A3), and learning—policy tuning, Basal Ganglia gates, and Amygdala sensitivity all consume AEF’s output so that the next intent can be gated or routed differently.
RFS can store execution state and telemetry when persistence or audit is required, so that runs are traceable and reproducible alongside the rest of the system’s memory and state.