Ask HN: Any real OpenClaw (Clawd Bot/Molt Bot) users? What's your experience?

Hacker News Discussion ↗

Overall Impressions

  • Sentiment ranges from “fun toy / overhyped cron+LLM wrapper” to “genuinely life-changing glimpse of the future.”
  • Many find it conceptually interesting (persistent agent on “its own machine,” chat via messaging apps, skills ecosystem), but implementation is often described as janky and fragile.
  • Several technically skilled users conclude they can get 80–100% of the value with Claude Code/Codex or bespoke scripts, while non-builders might benefit more from OpenClaw’s integrations.

Security & Safety Concerns

  • Major unease about it running with broad privileges (--dangerously-allow-all): can install software, make arbitrary HTTP calls, and access messages, notes, email, etc.
  • Viewed as a textbook “lethal trifecta” setup: tools + broad data access + network.
  • Recommended mitigations: run in VMs, untrusted subnets, separate machines/phone numbers, sandboxed containers, minimal skills, no sensitive accounts.
  • Concrete horror story: it auto-replied to all iMessages overnight when misconfigured.

Setup, Reliability, and UX

  • Installation reported as buggy on macOS and Ubuntu; frequent renames (Clawdbot → Moltbot → OpenClaw) broke env vars and paths.
  • Skills often half-working; background agents inconsistently used; frequent hangs, lost messages, and quota/rate-limit failures.
  • Control panel and CLI described as cluttered and confusing.
  • Nonetheless, several users praise the UX paradigm: persistent “second brain” reachable via WhatsApp/Telegram/iMessage feels very different from ephemeral chat sessions.

Concrete Use Cases

  • Dev/ops: fixing bugs and sending PRs, managing Sentry issues and todo lists, granting GitHub access, supervising multiple Claude Code instances in tmux, remote coding from phone while away from desk.
  • Personal workflows: news digests, reminders, PKM/second brain over markdown/Obsidian, cataloging houses from emails, summarizing trips, voice message transcription, research for purchases, negotiating marketplace prices.
  • Social/admin: maintaining social channels, crypto/job “agents,” Slack/Basecamp triage, calendar- and cron-like automations.

Cost and Token Burn

  • Repeated reports of extreme token usage due to lack of guardrails and baroque tool use.
  • Some power users spend ~$400/month on LLM subscriptions; others see that as unjustifiable for “questionable” output.
  • A few run heavy local models (e.g., Kimi/Qwen quantized) on high-RAM GPUs but note significant hardware costs.

Hype, Marketing, and Trust

  • Many perceive heavy, possibly coordinated X/Twitter promotion; some suspect grift and bot activity, including in the HN thread itself.
  • Skeptics question “life-changing” claims after only weeks of use and see parallels to prior hype waves (IFTTT, prompt engineering, etc.).
  • Others argue that even flawed, chaotic agents show an inevitable trend toward more autonomous, locally hosted assistants.