Where Hermes leads
- Channel breadth: many messaging adapters claimed
- Deployment shapes: Docker, VPS, remote backends
- Public community skills: open hub culture
Self-hosted agents people actually evaluate: Smartopol, Hermes, OpenClaw, NanoClaw, IronClaw, OpenJarvis, plus coding agents such as Claude Code and Grok Build. Where Smartopol leads, where others lead, and what is a different category entirely.
free to use · closed source by design · benchmarks you can run with smartopol bench
| capability | smartopol | hermes | openclaw | nanoclaw | ironclaw | openjarvis |
|---|---|---|---|---|---|---|
| Self-hosted | yes | yes | yes | yes | yes | yes |
| Nightly reflection | cortex dreaming | skill rewrites | ||||
| Hierarchical memory | Memory Pyramid | keyword / notes | basic | basic | encrypted KV | basic |
| Cross-channel identity | pairing codes | partial | per-channel | |||
| Channels (shipped) | Discord · TG · Web · WS · Terminal | many claimed | many claimed | few | limited | local |
| Model-agnostic | Grok · Claude · GPT · Qwen · Kimi · Ollama | Claude-first | Claude-first | Claude-first | Claude-first | Claude-first |
| Sandboxed plugins | signed, isolated | scripts | skills hub | basic | basic | basic |
| Memory write scanning | threat patterns + fence | partial | public CVEs 2026 | yes | ||
| OS sandbox | kernel-enforced | claimed | weak defaults | container | zero-trust | |
| Audit log | chained | claimed | ||||
| Self-update | signed binary | pip -U | manual | manual | manual | |
| Multi-user RBAC | yes | per-agent | yes | |||
| Published benchmarks | smartopol bench | |||||
| License | free to use · closed source | MIT | MIT | MIT | commercial | MIT |
| Pricing | free to use | free | free | free | paid | free |
| Commercial support | yes | yes |
Closest positioning rival: autonomous agent on your server that remembers and grows over time. Same category as Smartopol.
smartopol benchBoth free to use in spirit; Smartopol is closed source, Hermes is MIT.
Pick Hermes for open community + channel sprawl. Pick Smartopol for memory depth + operator hardening.
Strong on channels and skill marketplace. Weaker on default security posture.
If you want maximum marketplace size and accept marketplace risk, OpenClaw feels richer. If you want a smaller, sandboxed, operator-grade default, Smartopol is the point.
Minimal, auditable, security-focused alternative to larger OpenClaw-class stacks.
NanoClaw optimizes for smallest auditable surface. Smartopol optimizes for capable private agent with responsible defaults. Different products.
Enterprise zero-trust framing. Paid product path. Compliance vocabulary.
Local-first personal agent framing. Earlier-stage than the rest of this list. Smartopol is also local-first, and adds Memory Pyramid, channels, plugins, multi-user, and a published bench path.
Not substitutes. Typical setup: Smartopol as the long-lived agent; Claude Code (or Grok Build) as a focused coding hand.
Grok Build is xAI's coding agent product (Grok is the model family).Qwen Code and Kimi Code are coding agent CLIs. Aider and similar tools are repo-first editors.
Smartopol Agent is a general-purpose companion with memory, channels, scheduling, and multi-user service. It can code, and it can also keep life and ops context across weeks. Memory Pyramid is especially useful for long Grok Build sessions that need context across restarts.
If you only want a coding agent, use Grok Build, Qwen Code, Kimi Code, Claude Code, or Aider. If you want one long-lived agent that can use those as tools, use Smartopol.
smartopol bench after install.Invitations go out in waves. Free to use. Closed source by design.
No nurture sequence.