Self-host vs managed Hermes
Own the box, or operate from a control plane? Hours, failure modes, and a clear decision guide for hosted Hermes on SEAOTTER.
Self-host Hermes if you want the box. Use SEAOTTER if you want always-on Hermes without owning install, process supervision, TLS, and SSH recovery. Both paths work. Hermes stays open-source either way.
If Hermes is going into production—or onto a client retainer—you hit that fork fast. This is the tradeoff sheet we use.
Two jobs
Self-host Hermes means you own the install, the host, process supervision, secrets on the box, backups, and recovery. Maximum control. Zero vendor dependency for the runtime.
Managed Hermes on SEAOTTER means the control plane owns provision, isolation, TLS, backups, and lifecycle. Hermes remains Hermes. You create, poll, and run—pause, restart, restore, credentials, secrets, logs—without SSH into a VPS.
They solve different problems.
Where the hours go
| Workstream | Self-host | On SEAOTTER |
|---|---|---|
| First agent up | Installer + host + wiring | POST → provisioning → poll stages → running (typically a few minutes) |
| Keeping it up | Your supervisor, your restarts | Start / stop / restart APIs; dashboard when you need it |
| Secrets & credentials | On-box practices you invent | Reveal / rotate credentials; upsert secrets into Secret Manager |
| Second / Nth environment | Another VPS or VM story | Another isolated agent—own namespace, own quotas |
| Failure at 2 a.m. | SSH, dig, hope | Restart, restore from snapshot, or reprovision—hostname stays stable on restore |
| Security boundary | Whatever you harden | Per-agent namespace, NetworkPolicy, PSA restricted, gVisor on new agents |
First agent up
Self-host: Installer + host + wiring
On SEAOTTER: POST → provisioning → poll stages → running (typically a few minutes)
Keeping it up
Self-host: Your supervisor, your restarts
On SEAOTTER: Start / stop / restart APIs; dashboard when you need it
Secrets & credentials
Self-host: On-box practices you invent
On SEAOTTER: Reveal / rotate credentials; upsert secrets into Secret Manager
Second / Nth environment
Self-host: Another VPS or VM story
On SEAOTTER: Another isolated agent—own namespace, own quotas
Failure at 2 a.m.
Self-host: SSH, dig, hope
On SEAOTTER: Restart, restore from snapshot, or reprovision—hostname stays stable on restore
Security boundary
Self-host: Whatever you harden
On SEAOTTER: Per-agent namespace, NetworkPolicy, PSA restricted, gVisor on new agents
Self-hosting
- Installers and cloud accounts you keep watching
- SSH into boxes when something flakes
- Weekends lost to “just one more deploy”
- Another VPS for every client or team
On SEAOTTER
- Managed control plane—create, poll, run
- Isolated agents with lifecycle APIs
- Pause, restart, restore, secrets, logs—no SSH
- The agent keeps working when you close the laptop
Failure modes to design for
Self-host modes you will meet
- Host sleeps, disk fills, or the process exits—cron and “always-on” assistants go quiet.
- Credential sprawl across boxes; rotation is manual and easy to postpone.
- Client or team isolation defaults to “another machine.”
- You are on-call for every flake, even when Hermes itself is fine.
Managed-plane modes
- Agents pin to a Google Cloud region. Iowa (us-central1) is live. Other regions are coming soon.
- Deep OS customization and exotic networking may not fit the plane.
- Trial is 7 days with no credit card (one agent). Paid is $99 per agent / month. Custom is book a call. See pricing.
Knowing both lists is how you choose with confidence.
How the plane works
When you create an agent, SEAOTTER walks readiness stages in order:
namespace_created → helm_installed → pod_ready → sandbox_verified → dns_resolvable → tls_valid → wizard_reachable
New agents fail closed if the Ready pod is not under gVisor (sandbox_verified). Each tenant lives in agent-{id} with ingress only from the shared edge and egress limited to DNS + HTTPS. Secrets you upsert are stored as Secret Manager references—not pasted into a shared password doc.
Restore scales the workload down, rebuilds the Hermes home PVC from a VolumeSnapshot, and scales back up. The agent hostname does not change. That is the difference between “we have backups” and “you can operate at 2 a.m. without SSH.”
Regions and availability posture: Regions · Trust · Status.
Decision heuristics
Choose self-host when
- One personal operator, hardware you already run 24/7, and you want full control.
- You are learning Hermes by operating the stack end to end.
- Networking or data-residency constraints force you onto iron you control.
Choose managed when
- Always-on is part of the deliverable—standing assistant, scheduled research, client-facing agents.
- You need pause / restart / restore / secrets without SSH.
- You expect more than one isolated environment without multiplying VPS ops.
Price the plane against your nights
Plans cover compute, backups, and the control plane. Compare list price to boxes + your real calendar cost—then start the 7-day trial if the managed path wins.