AetherGrid
Open-source AI workload orchestrator, lease-based execution
1 follower
Open-source AI workload orchestrator, lease-based execution
1 follower
What happens when an AI workload loses its worker? AetherGrid keeps distributed jobs under control with intelligent scheduling, coordinated workers, and time-bound job ownership. An open-source control plane built for teams that want to orchestrate AI workloads across the compute they already own.
This is the 2nd launch from AetherGrid. View more
AetherGrid
Launching today
AetherGrid is an open-source control plane for distributed AI workloads. It schedules jobs across your existing compute, coordinates workers, and ensures each job has one owner at a time through time-bound leases, even during failures. Self-host it and orchestrate AI workloads on your own infrastructure.
Free
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where you’re already building
Promoted
What happens to your AI workload when the worker executing it suddenly disappears?
Does another worker take over? Who owns that job? Can two workers accidentally execute the same work? And how do you know what happened?
These are the kinds of problems that led me to build AetherGrid.
Hey Product Hunt! I'm Wycliffe Rotich, Founder and CEO of AetherGrid.
I've spent a significant amount of time building an open-source control plane for distributed AI workloads, with one central idea: a job should have one owner at a time, even when infrastructure fails.
AetherGrid coordinates workloads across the compute you already own. It brings together:
Multidimensional scheduling: Places workloads based on available compute resources.
Worker coordination: Manages distributed workers and their execution lifecycle.
Lease-based ownership: Uses time-bound leases to establish who owns a job.
Tenant-scoped access: Supports API key issuance for named tenants.
Execution reliability: Handles worker and process failures with predictable ownership and recovery behavior.
But here's something I want to make clear.
I didn't build AetherGrid to give the world another dashboard or wrap an existing orchestration tool in a new interface. I wanted to build the actual control plane: the scheduling, coordination and ownership mechanisms that determine how distributed workloads get executed.
The project is built in the open. You can inspect the implementation, read the Architecture Decision Records, run the tests, challenge the design and see how the pieces fit together.
Now I'd like to invite you to do something beyond simply looking at the launch.
Star the repository if you believe this is worth building.
Fork AetherGrid if you want to experiment, extend it or explore how it could fit into your infrastructure.
Run it and tell me what works, what doesn't, and where you think it needs to improve.
Challenge the engineering decisions. I'm particularly interested in feedback from people working with distributed systems, GPU infrastructure, AI workloads, platform engineering and cloud computing.
If you're operating compute infrastructure and have a workload orchestration problem, I'd also be interested in discussing a real technical evaluation.
This project has been built through a lot of engineering, testing and deliberate design decisions. Now I want more engineers to get their hands on it.
Explore AetherGrid:
GitHub: https://github.com/wycliffRotich-dev/aethergrid
Website: https://www.aethergrid.co.ke
I'm Wycliffe, building AetherGrid from Kenya, and I'm excited to see what this community does with it.
Let's build something useful.