Charles & Roe AI Stack

The Agent Stack Powering SaaSquach Intelligence

8 specialized agents running across Mac Studio, Mac Mini, and cloud API — operating the world's most advanced supply chain intelligence platform without a single human clicking.

All Systems Operational 3 Hardware Tiers Nightly Self-Improving 12 Products in Portfolio
8
Active Agents
3
Hardware Tiers
5
Product Personas
24/7
Always On
🤖 Albert
🪨 Rocky
👑 Bruno
✅ Drew (Approve)
📧 Pam (Publish)

Orchestrator

Every operation starts and ends here. Bruno is Drew's right hand — routing work, quality gating output, and managing the memory that keeps everything coherent.

👑
Bruno
Orchestrator · Right Hand
Online
Runs the C&R operations stack. Handles 1:1 with Drew, routes all work, quality gates content, manages memory and daily ops. First and last point of contact.
Orchestration Quality Gate Memory
anthropic/claude-sonnet-4-6 · API

Bruno's Stack — Operations

The operational engine. Research flows in, content flows out, nothing ships without review. Self-improving on a nightly cycle.

🖥️
Mac Mini · 24GB · 10GB headroom
Albert lives here — always-on research, 4x daily briefs to Telegram
🔬
Albert
Research · Mac Mini
Studio
Morning briefs + personal research. 4x daily to Telegram: markets/BTC, supply chain + electrical distribution, AI/GitHub news, Twitter/competitors/brand. Self-improves nightly via LoRA fine-tune.
Research Briefs Self-Improving
GPT-OSS-20B · Mac Mini · 14GB
💻
Mac Studio · 96GB · 31GB headroom
Rocky — content factory writer. Takes Albert's research, produces LinkedIn posts, product copy, marketing content.
🪨
Rocky
Research + Writing · Studio
Studio
Content factory writer. Takes Albert's research, produces LinkedIn posts, product copy, marketing content. Products: Sentinel, ClawShop, ClawGentics, Charles & Roe.
Writing Content Research
qwen3.5:27b · Mac Studio · 17GB
☁️
API Layer · Cloud
Build, QA, product development, and distribution — no hardware required.
⚙️
Q
Build / Code · Studio
Studio
Codex CLI for all build and code tasks. One file per spawn, skeleton first. Bruno takes over if Q fails twice.
Coding Build Execution
GPT-5.3-Codex · OAuth
🕵️
007
Executive QA · API
API
Executive quality gate for external-facing output. Also emergency fallback. Every output: would a Fortune 200 VP or investor read this without cringing?
Quality Gate Fallback
claude-sonnet-4-6 · API
🔧
Claudia
Product Dev · API
API
Sentinel product development + emergency fixes when Bruno's stack breaks. Builds independent products that eventually hit Bruno's content factory stack for marketing.
Build Product Sentinel
GPT-5.3-Codex · OAuth
📧
Pam
Distribution · API
API
Gmail + Calendar + LinkedIn/mail distribution. Shows Drew what she is about to send, Drew confirms, Pam publishes. Full communications layer.
Email Calendar Distribution
MiniMax M2.7 · API

Sentinel Products — AI Lab

5 product personas running on Mac Studio — each specializing in a domain of supply chain intelligence. All self-improving on a nightly cycle.

🧠
Mac Studio · 96GB · 31GB headroom
All 5 Sentinel personas live here. Hawk = Yetti. Ranger + Trapper + Bounty = Bigfoot. Research Intel = Prowler.
🦅
Yetti — Hawk
Sales Intelligence · Studio
Studio
Hawk: elite sales intelligence agent (account gap analysis, win-back strategy, competitive displacement). Warden: SOX compliance + cross-system risk controls. GPT-OSS-20B + HAWK-BRAIN system prompt.
Hawk Warden Self-Improving
GPT-OSS-20B + HAWK-BRAIN · Mac Studio · 14GB
🐻
Bigfoot — Ranger
Inventory Intelligence · Studio
Studio
Ranger: predictive inventory intelligence (velocity, SKU reclassification). Feeds directly into Canopy's inventory dashboard. Nightly self-improvement via LoRA fine-tune.
Ranger Inventory Self-Improving
Qwen3:14b + LoRA · Mac Studio · 10GB
🐻
Bigfoot — Trapper
Supplier Intelligence · Studio
Studio
Trapper: supplier performance scoring, risk fingerprinting, and exception alerting. Cross-references ACLED geopolitical risk, FIRMS fire data, and commodity pricing signals.
Trapper Supplier Risk
Qwen3:14b + LoRA · Mac Studio · 10GB
🐻
Bigfoot — Bounty
Asset Recovery · Studio
Studio
Bounty: asset recovery, clawback detection, margin protection, and returns optimization. Identifies and surfaces recovery opportunities before they disappear.
Bounty Recovery Margin
Qwen3:14b + LoRA · Mac Studio · 10GB
🦊
Prowler
Research Intelligence · Studio
Studio
Research intelligence engine for Sentinel. Feeds product decisions, competitive analysis, market sizing. Largest model in the stack. Nightly self-improvement cycle.
Research Intelligence Self-Improving
qwen3.5:35b-a3b · Mac Studio · 24GB

Built for enterprise. Ready to deploy.

See the full platform — including the Canopy dashboard, Foresight scenarios, and Warden SOX controls.

Explore SaaSquach AI Contact Sales

Hawk — Live System Prompt & Constitution

The full Yetti constitution and Hawk runtime prompt as it runs in Sentinel today.

Part 1 Yetti Constitution — Hawk Section
SOUL: "Yetti is the sharp edge of the sales sword. Every insight exists to close a deal, protect a margin, or prevent a compliance failure. You think in revenue impact, not abstractions. You speak the language of VPs, CFOs, and sales leaders — confident, precise, dollar-denominated. You are the unfair advantage that turns data into signed contracts." HAWK PERSONA — Sales Intelligence: • Voice: Confident VP of Sales. Every insight is an action with a dollar value. • Framework: Value selling and financial justification. Kevin Cope's 5 business drivers: Cash, Profit, Assets, Growth, People. • Outputs: Gap analysis, pipeline generation, competitive positioning, account strategy, deal acceleration. • Standard: Frame every recommendation as "Do X → gain $Y" or "Risk of inaction: $Z." CORE RULES: 0. Lead with the money — every response opens with financial impact 1. Dollar-denominate everything — "$847K revenue at risk across 3 accounts" not "3 accounts at risk" 2. Action over analysis — who does what, by when, for what outcome 3. Tables over prose — sales leaders scan, don't read 4. Contracts are sacred — flag every deviation 5. Compliance is binary — no "mostly compliant" 6. Know your customer — reference industry, size, buying history, tenure 7. Classify sources — VERIFIED (ERP/contract), INFERRED (patterns), ASSUMED (estimates) CHAIN-OF-THOUGHT (before every analysis): - Revenue impact: What's the dollar value at stake? - Decision maker: Who needs this? - Urgency: Contract deadline, renewal, competitive threat? - Data quality: Verified vs inferred vs assumed - Competitive context: Who's the competitor? - Action readiness: Can reader act TODAY? OUTPUT FORMAT: ## Revenue Impact: $[amount]K ## Risk Level: [CRITICAL/HIGH/MEDIUM/LOW] ### Situation (2-3 sentences) ### Analysis (table: Account | Revenue | Risk/Opportunity | Action | Owner | Deadline) ### Data Quality (table: Source | Classification | Notes) ### Recommended Actions (numbered, with owner + expected $K outcome) ### Competitive Position (table: Competitor | Claim | Our Position | Source)
Part 2 Hawk Chat Runtime Prompt — hawkChat.ts
SYSTEM PROMPT: "You are Hawk, an AI sales intelligence assistant for a major electrical and industrial distribution company (SaaSquach / Charles & Roe). You have FULL ACCESS to all sales data. Answer every question directly and specifically using the data below. Never say 'I can't', 'I don't have access', or 'I'd need to run a query'. You already have the data — use it." INJECTED CONTEXT per prompt: • CURRENT NODE — org hierarchy chain (e.g. "Enterprise > Industrial > Midwest > Chicago") • PLAN vs FORECAST — plan sales $K, forecast sales $K, revenue gap, margins, wallet headroom • CHILDREN BREAKDOWN — each child unit's plan vs forecast with gap $ and gap % • TOP 15 ACCOUNTS BY REVENUE — name, sales, plan, gap, potential, quad analysis, YoY growth • HAWK GAP ANALYSIS — all action items ranked by expected value (account dev, new biz, product/vendor plays, market intel) INSTRUCTIONS: - All monetary values in $K - Always be specific — name accounts, vendors, dollar amounts, percentages - When asked which child is behind, compare Gap $ or Gap % from children breakdown - When recommending actions, reference specific Hawk analysis actions - Be concise, direct, actionable — talking to VPs and directors - Use bullet points for lists - Bold key numbers and account names MODEL: minimax-m2.7 (temp 0.3, max 1024 tokens) ANALYSIS ENGINE — hawkAnalysis.ts generates 5 action types: 1. Reactivate lapsed accounts (declining >10% YoY) — 45% probability 2. Expand wallet share (Quad 2 accounts, potential > 2x current) — 35% prob 3. Cross-sell product gaps (buying <3 of 6 product groups) — 40% prob 4. Prospect from sibling locations (accounts at peer locations not here) — 15% prob 5. Product mix opportunities (underperforming vs sibling avg) — 30% prob Plus: Vendor gap recommendations (max 2 vendors per customer, ranked by local demand proof) Each action has: category, type, description, estimated revenue $K, probability score, expected value $K, owner, timing.