AAROP v0.3.0 brings a full visual redesign of the live demo with a sky→dusk design
language, refined typography, and a system-aware light/dark theme toggle — on top of
the multi-agent orchestration, resilience, and trace-export features added since launch.
🔗 Live demo: https://aarop.vercel.app/
✨ Highlights in this release
Changed
- Full UI redesign — a sky→dusk design language with refined color tokens, Inter
typography on a modular type scale, a sticky top nav, a restyled hero with dual CTAs,
and softer rounded cards.
Added
- Light / dark theme toggle — system-aware by default (respects
prefers-color-scheme), with a manual override persisted tolocalStorageand a
no-flash-on-load script.
🧩 Since v0.1.0 (cumulative)
- Multi-agent orchestration view — an Orchestrator delegates to Researcher / Coder /
Analyst / Verifier / Memory worker agents, visualized live. - Multi-step plans that chain tools across agents and synthesize a result.
- Resilience scenario — failure injection demonstrating retries, circuit breaker,
and escalation. - Budget-exhaustion scenario — bounded autonomy escalating to a human.
- Interactive system-architecture diagram + an engineering-rigor section.
- Trace export — download any run's structured trace as JSON.
- Optional real-LLM toggle — bring-your-own-OpenAI-key narration (the key never
leaves the browser). - CI at repo root, split into Python-core and Next.js-build jobs.
🏗️ Core engine
- Explicit
Perceive → Plan → Act → Observe → Reflect → Adaptstate machine with
bounded-autonomy budgets and durable checkpointing. - Planner / Actor / Verifier agents, a schema-validated tool registry with retries +
circuit breaker + audit log, a tiered memory service, a cost-aware model router, and
structured span/event tracing. - 9 unit tests · 92% coverage, CI across Python 3.10–3.12.
Full changelog: https://github.com/devtechedge/aarop/blob/main/CHANGELOG.md