Show HN: DenchClaw – Local CRM on Top of OpenClaw

Product scope and positioning

• Presented as a local-first CRM layer on top of OpenClaw, but repeatedly reframed as a general “everything app” for Mac: natural-language control over browser, files, notes, calendar, slides, code, and research.
• CRM is described as just one app on this platform (accounts, contacts, pipelines, etc.), with the long‑term vision of “database tables + skills + agent.”
• Some find the messaging confusing: is it primarily a CRM, a “Cursor for your Mac,” or a generic agent workspace?

Use cases and value

• Reported everyday uses: aggregating meeting notes, cross‑referencing tasks, drafting follow‑ups, reorganizing research folders, building slides, coding, and searching Notion/markdown knowledge bases.
• For outreach, users still need to edit “robotic” drafts; commenters say the real time savings are in data gathering, enrichment, and context assembly, not text generation itself.
• Several commenters prefer simpler setups (plain‑text CRM in Obsidian/Logseq) for solo work.

Security and privacy concerns

• Strong and repeated warnings that OpenClaw‑style agents are “barely secure enough to sandbox,” with huge attack surfaces due to full system and browser access.
• Copying the Chrome profile so the agent can “see what you see” is seen by many as extremely risky: exposes credentials, private messages, financial data, and sensitive personal information; prompt injection and hallucinations remain unresolved.
• Commenters worry about state actors and cybercrime targeting such tools, likening current enthusiasm to past unsafe tech fads.
• Local‑only data storage is not viewed as sufficient mitigation; the core fear is high‑privilege, automated control of real accounts, devices, and infrastructure.

Spam and outreach debate

• Many see agent‑driven CRM as an “AI spam machine” for email and LinkedIn, worsening an already bad cold‑outreach ecosystem.
• Others argue that sales/prospecting will inevitably adopt these tools and that outreach is only a small fraction of potential use; however, there is no practical way to “enforce” non‑spammy uses in open source.

Architecture, models, and cost

• Uses DuckDB for local‑first data; some question why not SQLite and why not just files, given many fields are strings.
• Some suggest focusing agent use on read‑only enrichment scripts that output CSVs for manual CRM import to reduce risk.
• Small local models are reported as too weak for complex tasks; recommended practice is using high‑end cloud models (e.g., Claude variants), sometimes with escalation strategies, but costs can become substantial without careful limits.

Ecosystem, hype, and platform choices

• Concerns that GitHub stars are a poor signal in the current hype cycle; some suspect aggressive promotion.
• Building atop OpenClaw is criticized due to perceived “vibe‑coded” quality and security issues, though others are still excited about the general “skills in a filesystem” direction.
• Questions about why people run this on Mac minis rather than VMs; answers center on iMessage integration and cultural “Mac mini as ritual hardware.”