Splendor Kernel
Persistent execution, permission checks and replay for AI agents, with a Rust core and Python and TypeScript clients.
Open-source agent runtimeJan 2026 — Present
Overview
An open-source Rust runtime that lets AI agents retain state, resume work and check permissions before taking action.
Technical work
- Implemented scheduling and resumable loops, with SQLite state graphs, snapshots and integrity-chained execution traces for replay.
- Added an action gateway that checks permissions and verification rules before tool execution, alongside typed messaging, scoped delegation and per-agent isolation.
- Extended the runtime with signed work orders, state handoff, approvals, revocable policies and circuit breakers. Edge-device contracts cover offline operation, reconnect trace sync and bounded robotics adapters with simulation harnesses.
- Published a CLI, OpenAPI service, Python and TypeScript clients, reference examples and documentation. The Rust core enforces runtime rules independently of the model calling it.