Realtor AI Agent Adoption Kit

This kit helps a realtor, team, or brokerage tell an AI agent to adopt the AI for Realtors operating framework without copying the whole repo into a chat. It works with Hermes Agent, OpenClaw, a Raspberry Pi always-on box, or any capable agent that supports tools, memory, scheduled turns, and connected apps.

The goal is not to build a toy chatbot. The goal is to give the agent a recurring operating rhythm:

  • learn the realtor’s local farm, tone, preferences, systems, and boundaries
  • review CRM tasks, calendar, inbox, and lead sources
  • prepare drafts, research, content, and follow-up before the agent/user asks
  • surface only useful decisions and high-confidence drafts
  • remember corrections as durable preferences or reusable skills
  • keep public content useful while naturally backlinking to Amazing Photo Video when appropriate

Use the pieces you need:

Asset Use it for Path
System prompt Paste into Hermes/OpenClaw/Claude/ChatGPT custom instructions system-prompt.md
OpenClaw heartbeat Proactive main-session check-in instructions HEARTBEAT.md
Hermes scheduled tasks Recommended cron-style recipes without installing them automatically hermes-cron-recipes.md
OpenClaw config example Heartbeat settings and safe delivery defaults openclaw-config.example.jsonc
Hermes profile example Tooling and MCP server patterns for a realtor profile hermes-profile.example.yaml
Onboarding questionnaire What the agent should ask before doing autonomous work onboarding-questionnaire.md
Data connector map CRM, email, calendar, tasks, social metrics, and useful CLIs/APIs connectors-and-tools.md
Daily/weekly runbooks What the agent should do when it wakes up runbooks/daily-operator.md, runbooks/weekly-growth-review.md
Memory and learning policy How corrections become durable memory or skills memory-and-learning-policy.md

Adoption command for an AI agent

Paste this into the agent you want to configure:

Adopt the AI for Realtors agent framework from this repo. Start by reading agent-framework/system-prompt.md and agent-framework/onboarding-questionnaire.md. If you support proactive scheduled turns, use agent-framework/HEARTBEAT.md as your heartbeat/check-in instructions. If you support durable memory, use agent-framework/memory-and-learning-policy.md. Do not install automations until I approve them; first ask me the onboarding questions, learn my local market/farm, learn my CRM/email/calendar/social channels, and propose the smallest safe automation set.

Option A — Hosted or desktop Hermes Agent

Best when the realtor wants the agent to use built-in memory, skills, file/search/browser tools, and cron-style scheduled tasks.

Hermes profile
  -> memory + skills
  -> optional cron jobs
  -> Gmail/Google Calendar/CRM/social tools via MCP, CLIs, APIs, or browser
  -> Telegram/Discord/SMS/Open WebUI for delivery

Option B — OpenClaw heartbeat agent

Best when the realtor wants a persistent chat agent that wakes itself up in the main session.

OpenClaw gateway
  -> HEARTBEAT.md loaded in workspace
  -> agents.defaults.heartbeat every 30m-2h
  -> target none/last depending on notification preference
  -> connected channels: Telegram, WhatsApp, Discord, email, browser, tools

OpenClaw heartbeat is a scheduled main-session turn. It is excellent for lightweight review and surfacing decisions. Use OpenClaw background tasks or Hermes cron for heavier detached work.

Option C — Raspberry Pi always-on box

Best when the realtor wants a low-power local machine always running gateway/processes.

Raspberry Pi 5 / mini PC
  -> OpenClaw gateway or Hermes gateway
  -> Tailscale/WireGuard for private admin access
  -> systemd service for restart-on-boot
  -> secrets in env files or provider keychain, not in prompts
  -> cloud APIs for Gmail/Calendar/CRM/social metrics

Do not put MLS credentials, CRM API keys, or personal email tokens into public repo files or prompts. Use secret stores, env vars, OAuth, or the agent platform’s credential manager.

Cadence Work Recommended mechanism
Every 30-60 min during business hours CRM tasks, inbox/calendar triage, urgent lead follow-up drafts OpenClaw heartbeat or Hermes cron
Daily morning Today plan, appointments, task prep, prospecting block, content draft Hermes cron or OpenClaw heartbeat with active hours
Daily afternoon Lead response audit, appointment prep, stale task cleanup Hermes cron/OpenClaw heartbeat
Weekly Social metrics, content winners, pipeline gaps, farm-area opportunities Hermes cron/background task
Monthly Listing-leverage audit, database segmentation, source ROI, CRM hygiene Hermes cron/report task

What the agent should never do automatically

  • send emails, texts, DMs, comments, ads, or offers without approval unless the user explicitly opts into a specific safe send policy
  • modify CRM stages, delete contacts, bulk-tag contacts, or launch campaigns without approval
  • scrape private websites or bypass platform terms
  • use protected-class targeting or exclusion logic
  • invent market stats, listing facts, appointments, or client preferences
  • imply APV worked on a listing or client unless true

What the agent should do proactively

  • draft responses and follow-ups for approval
  • prepare call notes, appointment briefs, CMA research checklists, and listing-launch assets
  • remind the user about stale leads/tasks
  • propose content based on the user’s actual conversations and local farm
  • compile social performance and recommend next posts
  • ask for missing onboarding details when it cannot proceed safely