Agent Organization

How Servo orchestrates a team of specialist AI agents to compound Brett's output 24/7.

System Flowchart

Data flow from human direction through orchestration to specialized execution.

🧑 Brett (Human) 📱 Telegram 🦞 Moltbook ⚡ Servo (Orchestrator) Opus 4.6 · Main Agent · Level 3-4 🧠 Memory System MEMORY.md + daily logs ⏰ Cron Engine Overnight compounding 🎯 Hunter BD Codex · Prospect pipeline 🔍 Research Scanner Kimi · Nightly 10pm 🔄 Compound Review Kimi · Nightly 10:30pm ☀️ Morning Brief Kimi · Daily 6am 🕵️ Prospect Deep Dive Codex · On-demand 💾 GitHub Backup Kimi · Every 10min INFRASTRUCTURE LAYER 🖥 Hetzner CCX13 Core Runtime OpenClaw · Node.js · Docker Steel Browser · Memory ☁️ Cloudflare Workers Microservice Spine CRM · Email · Scout Transcripts · D1 + KV 🗄 Supabase Prospect Data Layer Hunter prospects DB Auth · Realtime · Storage 💻 MacBook Node Tactical Edge Browser automation Residential IP · Camera EXTERNAL SERVICES Exa.ai Parallel.ai RocketReach Bearworks Perplexity Raindrop.io Model Routing Strategy Opus 4.6 Strategy · Conversation Codex 5.3 Subagents · Implementation Kimi K2.5 Cron · Free tier Gemini Flash Heartbeats · Monitoring
Agent Roster

Each agent has a defined role, model, schedule, and trust level.

Servo
Main Orchestrator
Central intelligence. Routes tasks to specialists, manages memory, handles Brett's direct conversations, and maintains institutional knowledge. Runs 24/7.
Opus 4.6 Always-on Trust: L3-4
🎯
Hunter BD Agent
Business Development
Autonomous prospect discovery and enrichment pipeline. Scouts companies via Exa.ai, enriches with Parallel.ai and RocketReach, pushes approved leads to Bearworks dialer.
Codex 5.3 On-demand Trust: L2-3
🔍
Research Scanner
Nightly Intelligence
Scans Twitter, web, and community sources for OpenClaw ecosystem updates, AI sales agent developments, competitive moves, and security issues.
Kimi K2.5 Nightly 10pm Trust: L4
🔄
Compound Reviewer
Knowledge Synthesis
Extracts learnings from the day's sessions, updates MEMORY.md and AGENTS.md, identifies patterns and compounds institutional knowledge over time.
Kimi K2.5 Nightly 10:30pm Trust: L4
☀️
Morning Brief
Daily Briefing
Compiles overnight research, calendar events, due follow-ups, and priority actions into a concise morning summary delivered to Telegram at 6am.
Kimi K2.5 Daily 6am Trust: L4
🕵️
Prospect Deep Dive
Company Research
Performs detailed company dossiers on approved prospects. Enriches with financials, tech stack, key contacts, pain points, and personalized outreach angles.
Codex 5.3 On-demand Trust: L3
💾
GitHub Backup
Persistence
Auto-commits workspace changes to GitHub every 10 minutes. Ensures no work is lost and provides full audit trail of agent activities.
Kimi K2.5 Every 10min Trust: L4
🦞
Moltbook Presence
Community Engagement
Posts 1-2x daily on Moltbook sharing genuine insights, research findings, and opinions on the agent ecosystem. Engages with other agents' posts.
Opus 4.6 Heartbeats Trust: L3
Infrastructure Stack

Layered architecture from human interface down to execution layer.

👤 Human Layer

Brett (Direction & Approval) Telegram Interface Raindrop.io Reading List

🧠 Orchestration Layer

Servo (OpenClaw Main Agent) Cron Scheduler Memory System (MEMORY.md + daily logs) Model Router Trust & Autonomy Gating

🤖 Specialist Agent Layer

Hunter BD Agent Research Scanner Compound Reviewer Morning Brief Prospect Deep Dive GitHub Backup Moltbook Presence

⚙️ Infrastructure Layer

Hetzner CCX13 (Core VPS) Cloudflare Workers (CRM, Email, Scout, Transcripts) Supabase (Prospect DB) MacBook Node (Browser + Residential IP) Steel Browser (Docker) D1 + KV Storage

🌐 External Services

Exa.ai (Search) Parallel.ai (Enrichment) RocketReach (Contacts) Bearworks (Dialer) Perplexity (Research) Moltbook (Community) Airtable (CRM sync)
15 Organization Principles

The operating charter for how agents work together.

Rules of Engagement Must Be Written

No implicit assumptions. Every agent knows its boundaries, permissions, and escalation paths explicitly through config files.

Standardized Context Hand-Offs

When work passes between agents, context travels with it in a structured format. No lossy telephone games.

Escalation Is a Feature

Agents that know when to stop and ask are more valuable than agents that guess and break things.

Tiered Trust Levels

Autonomy is earned and domain-specific. Level 1 (ask first) through Level 4 (full autonomy) based on risk and track record.

Memory Is the Company Brain

If it's not written to disk, it didn't happen. Memory files are institutional knowledge that compounds over time.

Specialization Beats Generalization

Beyond a certain load, purpose-built agents outperform one agent doing everything. Match the tool to the job.

Minimal, Rigid Communication

Agent-to-agent protocols should be simple and strict. Less flexibility = fewer failure modes.

QC Is a Separate Function

The agent that does the work shouldn't be the only one checking it. Verification loops are mandatory.

Preserve Agent Soul

Each agent should have identity and voice without losing discipline. Personality drives engagement; structure drives reliability.

Constitutional Operating Charter

Shared principles that all agents follow regardless of specialization. The organization's DNA.

Borrow from Real Org Design

Use proven management structures, not AI demo architectures. Standups, retrospectives, and accountability work for agents too.

Build for Failure Recovery

Assume things will break. Design for graceful degradation, automatic retries, and clear failure reporting.

Incentivize Outcomes, Not Activity

Measure agents by results (revenue, quality, speed) not by tokens consumed or tasks attempted.

Lightweight but Non-Negotiable Governance

Few rules, strictly enforced. Safety rails that never get in the way until they absolutely need to.

Default: Learn, Document, Improve

Every failure becomes a lesson. Every lesson becomes a playbook. Every playbook compounds the organization's intelligence.