inner stratus
Hi there,
I’ve been working on a project called Leion Roots and I’d love to get some feedback from people building similar systems.
What it is (short version):
It’s basically a “Middle OS” for AI workflows inside VS Code.
Instead of using agents as black boxes, the idea is to orchestrate everything through structured pipelines.
Core idea:
- AI is not the brain → it’s a tool
- The system (Leion) handles routing, execution, and control
- Everything is transparent, replayable, and deterministic when needed
What I just connected:
- OpenClaw running locally (multi-agent via Discord threads)
- Integrated into my workflow
- Each thread = isolated agent context
- Can generate tools, scripts, and workflows that are directly reusable
What Leion Roots currently does:
- Pipeline-based orchestration (like a graph OS)
- Multiple node types (AI, shell, HTTP, Git, memory, etc.)
- Full logging + replay of executions
- Strong human-in-the-loop control
- Runs both in VS Code and CLI
What I’m exploring now:
- Using OpenClaw as a router / worker, not just an agent
- Letting it generate its own tools inside a controlled repo
- Multi-model routing (local / Codex / Claude CLI style)
- Cost-aware + deterministic execution
I’m curious:
- Are some of you building similar “AI orchestration layers” instead of just agents?
- How do you handle routing / control / cost?
If you want to take a look or test it:
👉 https://github.com/Leion-wp/intent_router
Current dev branch codex/Windows
Happy to exchange ideas or even collaborate 🚀
