4-Agent AI System Configuration & Stabilization (OpenClaw) Project Overview We need an experienced developer to configure and stabilize a multi-agent AI system for a service provider company. The system runs locally on macOS using OpenClaw and integrates with OpenAI and Brave Search. The main goals are: Persistent memory (no session wipe) Strict role isolation between agents No task overlap or role confusion Secure, scalable architecture Current Setup OS: macOS (Mac mini) Platform: OpenClaw (local gateway/dashboard) APIs: OpenAI API (already connected) Brave Search API (already configured) Execution: Manual / on-demand (no automation yet) Existing agents: Infrastructure agent Finance/Organization agent Required Final Architecture (4 Agents) Each agent must have: Its own OpenClaw workspace Separate persistent memory Scoped tool access Strict permission boundaries 1) Marketing & Social Media Agent Responsibilities: Content calendars & social posts Ad campaigns & outreach Lead funnel analysis Restrictions: No access to financial or admin data 2) Training & Operations Agent Responsibilities: SOP & playbook management Technician onboarding & performance tracking Scheduling & operational dashboards Access: Read-only access to finance summaries (no raw banking data) 3) Admin Agent Responsibilities: Email triage & scheduling HR paperwork & compliance Cross-team coordination Access: Read-only access outside its own domain 4) Finance Agent Responsibilities: Daily / weekly / monthly reports Budgeting & forecasting Revenue reconciliation Tip & payout tracking Restrictions: No access to marketing or social media systems Memory & Data Requirements Persistent Memory (Mandatory) Separate memory per agent Use: SQLite or Postgres for structured data Chroma or LanceDB for vector memory No shared memory between agents Data to Ingest SOPs & process documents Policies from SOUL.md / USER.md Training & onboarding checklists Housecall Pro exports & KPIs Security & Permissions Separate OpenClaw workspaces Tool-level permission scoping Memory isolation at database & vector level No credential sharing between agents Scope of Work Re-configure existing agents Create and configure 2 additional agents Implement persistent memory for all agents Enforce strict role boundaries Validate behavior (no role leakage) Out of Scope UI/dashboard development Cloud deployment Advanced automation (cron jobs can be added later) Expected Outcome A stable, secure, and scalable 4-agent AI system where: Memory persists across sessions Each agent stays in its lane Sensitive data is protected System is ready for future automation Budget: $45 NZD Delivery: 3 days