Celistra vs HashiCorp Nomad
Nomad is for production workloads at small-to-medium scale: containers, raw binaries, VMs, with placement, allocation, fault tolerance. Celistra is for AI agent supervision and homelab process management — simpler, mobile-first, sandbox-by-default. Different scopes.
Architecture
| Nomad | Celistra | |
|---|---|---|
| Schedulers | Pluggable (system, batch, service, parameterized) | None (you say where it runs) |
| Workload types | Container, exec, raw_exec, VM, custom | OS processes (PTY) |
| Cluster size | 1–10,000+ nodes | 1–50 nodes (homelab/indie scale) |
| State store | Raft (HA) | Firestore (cloud) + per-daemon SQLite |
| Placement constraints | Yes (datacenter, OS, GPU, etc.) | Manual (you pick the host) |
| Mobile UX | None | Native iOS/Android |
| Sandbox per task | Container isolation if you use containers | macOS Seatbelt / Linux bwrap built-in |
| Audit log (tamper-evident) | Audit logging exists, not hash-chained by default | Hash-chained, hourly verified |
When to pick Nomad
- You're running production services with HA requirements
- You need automatic placement based on resource constraints
- You have 10+ nodes and care about scheduler efficiency
- You're already in HashiCorp's stack (Consul, Vault)
When to pick Celistra
- You're supervising AI agents on personal/dev machines
- You want a phone app for approvals
- Sandboxing is part of the threat model
- "Run 10 Claude Codes for an afternoon" is the use case
Use both
Nomad for production services on a server fleet; Celistra on developer / researcher boxes. They're at different parts of the stack.
FAQ
Can I run Nomad and Celistra on the same machine?
Yes. They don't fight — Nomad's tasks are usually long-lived containers; Celistra's workload is interactive PTYs. Different process namespaces in practice.
Does Celistra plan to add scheduling?
No. The 'just say which host' model is intentional. Once you want a scheduler, you've outgrown Celistra and Nomad is right.