I'm a 3rd-year software engineering student. For the past 10 days, I've been on a solo sprint to build the foundation for Loom, an open-source, event-driven OS for AI agents.
My frustration with current agent frameworks is that many are request/reply. They aren't built from the ground up for agents that need to continuously sense, reason, and act in the real world.
Loom is my attempt to solve this. It's event-driven, not request-driven.
The core is built in Rust (performance, safety, concurrency). It uses a gRPC bridge to connect to a polyglot ecosystem, starting with a Python SDK (loom-py).
The architecture (detailed in the README) is designed for production from day one, with features like:
An Event Bus with QoS levels (Realtime, Batched) and backpressure handling.
Native multi-agent collaboration (fanout/fanin, contract-net) via an Envelope metadata system.
An ActionBroker for unifying tools, including MCP (Model Context Protocol) integration.
I've gone as fast as I can alone. Now I need help to "go far."
As a student, I know I have massive blind spots. I'm posting this today to find:
Mentors: I'd be grateful for code reviews and architectural critiques from senior engineers.
Collaborators: If this vision resonates with you, I'd love to build it together.
Harsh Feedback: What's wrong with this design? What will fail?
The code is alpha, but the vision is documented. Thank you for taking a look.
Hi HN,
I'm a 3rd-year software engineering student. For the past 10 days, I've been on a solo sprint to build the foundation for Loom, an open-source, event-driven OS for AI agents.
Link: https://github.com/your-username/loom
My frustration with current agent frameworks is that many are request/reply. They aren't built from the ground up for agents that need to continuously sense, reason, and act in the real world.
Loom is my attempt to solve this. It's event-driven, not request-driven.
The core is built in Rust (performance, safety, concurrency). It uses a gRPC bridge to connect to a polyglot ecosystem, starting with a Python SDK (loom-py).
The architecture (detailed in the README) is designed for production from day one, with features like:
An Event Bus with QoS levels (Realtime, Batched) and backpressure handling.
Native multi-agent collaboration (fanout/fanin, contract-net) via an Envelope metadata system.
An ActionBroker for unifying tools, including MCP (Model Context Protocol) integration.
I've gone as fast as I can alone. Now I need help to "go far."
As a student, I know I have massive blind spots. I'm posting this today to find:
Mentors: I'd be grateful for code reviews and architectural critiques from senior engineers.
Collaborators: If this vision resonates with you, I'd love to build it together.
Harsh Feedback: What's wrong with this design? What will fail?
The code is alpha, but the vision is documented. Thank you for taking a look.