OpenRig

The Actual Problem: Your Agent Topology Dies When You Hit Reboot

Let's get the pitch out of the way. If you run two or three Claude Code or Codex sessions in parallel on a project, every restart nukes the entire setup. Each session is an island. The relationships, shared context, task assignments—all gone. OpenRig exists to fix exactly this.

The project bills itself as "Terraform for coding agents." One YAML file declares your entire multi-agent topology. One command boots the whole fleet. Persistent identity, shared memory, snapshot and restore—primitives that cloud infrastructure has had for a decade but nobody bothered to bring to the coding agent layer. The creator, mschwarz, put it plainly: the topology "survives real work instead of collapsing at reboot time."

This matters now because the industry is clearly moving toward multi-agent workflows. Anthropic launched Claude Managed Agents in April 2026. But in practice, if you're mixing Claude Code and Codex to build something, your orchestration is throwaway. OpenRig treats that orchestration as infrastructure worth persisting. That's a real insight, not a buzzword.

Context Domains: Not Just Process Management

At localhost:7433 you get a visual graph of your entire topology—clickable, inspectable, live. But the more interesting idea buried in the HN discussion is what mschwarz calls "Context Domains." The concept: by distributing a task across multiple agents, each working on a slice, you effectively extend the usable context window beyond what any single LLM session can hold.

This reframes what OpenRig is doing. It's not just spinning up processes and keeping them alive. It's structuring agent collaboration so that the whole topology can tackle problems that would overflow a single context window. Whether this actually works at scale is an open question, but the framing is compelling—and specific enough to evaluate.

How do agents talk to each other? The answer is deliberately mundane: tmux. Mschwarz called it "super simple but effective." For task handoffs and coordination, there's a stripped-down built-in protocol, intentionally kept minimal so a single human operator can inspect what's happening across the fleet.

Honest Trade-offs: No Isolation, No Pretense

This is where I give OpenRig credit for intellectual honesty. HN user NBenkovich asked the obvious follow-up: if you're using tmux, every agent shares the same network and filesystem. What about isolation?

Mschwarz's response: "OpenRig isn't trying to solve that problem yet." The project assumes you're running agents on a host you control—your Mac, a VPS—and that you want frictionless file system access across agents. That's the whole point of collaboration. It's a conscious trade-off: convenience and cohesion over sandboxing. For a solo developer running a personal fleet of coding agents locally, this is perfectly reasonable. If you need production-grade agent isolation, look elsewhere.

Another gap: no A2A protocol support. Mschwarz acknowledged "there's a ton of overlap" and called it a potential roadmap item, but it's not there today. Teams eyeing Google's A2A ecosystem should note this.

Who Should Care (And Who Shouldn't)

The HN thread got 6 points and 8 comments. That's early-stage traction at best, and mschwarz confirmed this is "the first version I'm actually inviting other agent-heavy developers to try." Don't expect a polished platform.

The ideal user is specific: you run 2–3 parallel coding agents daily, they need shared context and task handoffs, and you're tired of rebuilding that setup from scratch every session. You're one person managing a fleet, and you want to see what's happening at a glance.

Not for you if: you're building team-level agent infrastructure, you need container-level isolation, or you occasionally open a single Claude Code session to write some code. The entire abstraction is designed around "a single human operator managing a fleet of agents"—mschwarz's exact words.

Verdict: Right Problem, Very Early

OpenRig is one of those projects where someone clearly identified a gap that the big players haven't addressed yet. The idea that multi-agent coding sessions deserve the same infrastructure primitives as cloud deployments—persistent state, declarative config, snapshots—is sound. The execution is early and the communication layer (tmux) is charmingly low-tech.

The real question isn't whether OpenRig is good today. It's whether multi-agent development becomes a standard workflow in the next two years. If it does, someone needs to solve the persistence and orchestration problem, and OpenRig has a head start on the mental model. If multi-agent coding stays a niche power-user pattern, this stays a niche tool. Either way, it's a project worth watching if you're already living in the multi-agent future and tired of your topology dying every time you close your laptop.