Systems Research
Kernos OS
A dual-layer autonomous operating system treating AI agents as first-class kernel citizens rather than user-space applications. Combines a high-speed Go 1.23+ microkernel backend featuring high-dimensional decaying/reinforcing vector graph memory and self-healing task DAGs with a React 19 / TypeScript desktop windowing environment.
Engineering objective
Build an operating system substrate where cognitive agents operate as isolated kernel-level citizens coordinated by a self-healing microkernel, synaptic vector graph memory, and peer-to-peer event routing.
Architecture
Kernos OS is architected in two complementary tiers: a high-performance Go microkernel server managing system processes, self-healing task DAGs, and high-dimensional vector memory graphs, communicating with a modular React 19 / Vite desktop shell providing visual multi-agent workspaces, terminal pipelines, and hardware monitors.
- Agents as Kernel Citizens: Cognitive agents are managed as kernel-level processes with explicit scheduling and memory limits rather than unconstrained user applications.
- Self-Healing Graph Execution: When execution steps fail, the task engine dynamically races alternative recovery branches across Directed Acyclic Graphs.
- Synaptic Memory Dynamics: High-dimensional vector graphs decay inactive memories and reinforce frequently traversed associations.
- Go Microkernel Core: High-speed event coordinator and scheduler managing isolated agent sandboxes, system Undo, and capability policies (server/main.go).
Implementation evidence
Parallel Recovery Branching
OBSERVEDVerified via server/task_engine_test.go that racing parallel DAG recovery paths resolves tool execution stalls without manual intervention.
Synaptic Vector Graph Reinforcement
OBSERVEDVerified via server/vector_engine_test.go that repeated query activation dynamically boosts graph edge weights while unreferenced nodes decay.