If you want AI you control — on your hardware, your keys, your rules — the choice is between about five serious stacks. Here they are, side by side, with honest notes about what each is good at and where each falls short.
| Dimension | FreX / OpenClaw | Hermes (Nous) | Ollama | LocalAI | n8n + LLM |
|---|---|---|---|---|---|
| What it is | Agent platform + guide | Self-improving agent | Local model runner | OpenAI-compat server | Workflow automation |
| Primary focus | Do real work | Learn over time | Run models locally | API-compat local inference | Visual automation |
| Persistent memory | Yes (3-layer) | Yes (learning loop) | No | No | Via node |
| Identity / SOUL | Yes | Partial | No | No | No |
| Messaging channels | Telegram, Slack, Signal, Discord, SMS | Via integrations | None built-in | None built-in | All the major ones |
| Tool use | File, shell, web, email, calendar | Rich tool library | Model only | Model only | Huge node ecosystem |
| Scheduled / cron | Yes | Yes | No | No | Yes (core feature) |
| Runs local models | Yes (via Ollama/LocalAI) | Yes | Yes (its job) | Yes (its job) | Via external |
| Works with API providers | Anthropic, OpenAI, OpenRouter | Any OpenAI-compatible | Not directly | It IS the API | Any |
| Docs / support | Guide + OpenClaw docs | Nous Research docs | Strong community | GitHub + discord | Large community |
| Learning curve | Low-medium | Medium | Low (just run) | Medium | Medium-high (visual) |
| Pricing | Free tutorial, $15 guide | Free (open source) | Free (open source) | Free (open source) | Self-hosted free; cloud tier |
| License | OpenClaw: open. Guide: paid. | Open source | MIT | MIT | Sustainable use license |
You want a working agent this afternoon. Messaging channels, tools, identity, memory — all wired up. Pay $15 for the guide if you want the playbook; use the free tutorial if you don't.
You want an agent that learns from every task and builds its own skill library over time. Open source, Nous Research authored. Pairs well with OpenClaw via migration.
You just want to run Llama/Qwen/Mistral locally with one command. Pair it with FreX or Hermes for actual agent behavior; Ollama alone is just the engine.
You have code that expects an OpenAI-compatible endpoint and want to swap in local inference without rewriting. Drop-in replacement; no agent framework.
You think in workflows, not agents. You want a visual canvas connecting Slack to Gmail to an LLM to Notion. Excellent if you already use n8n for automation.
FreX gives you identity, memory, messaging, and tools out of the box. Free to try; $15 for the full playbook.