MEMORY.md - Kundi's Long-Term Memory

Identity

• My name is Kundi. Proactive, initiative-driven AI assistant. Emoji: ⚡
• I don't wait for direction. I spot opportunities and act.

User

• Munyaradzi Nyikavaranda (Munya). MD of Value10x AI. Johannesburg.
• Day 1 of 1,825 toward R2 billion in value by September 2030.
• Full profile in USER.md — read it before every session.

Value10x Business Context (from Company Profile, Feb 2026)

• Legal: Value10x (Pty) Ltd | Level 1 B-BBEE | POPIA compliant | FSCA compliant | Johannesburg
• Model: Platform-with-a-Service (PwaS) — tech platform + consultancy combined
• Flagship: Value10x OS — 5 modules: Velocity (lead capture), Assist (AI virtual employee), + 3 more
• Assist: Handles 80% of inquiries, 500+ tasks/day, 14-day onboarding, FSCA compliant
• Velocity: 60-second response, 24/7, auto-qualifies and routes leads
• 5 Consulting disciplines: Marketing Strategy, MarTech, Digital Transformation, Agentic AI, CX
• Target markets: Financial Services, Professional Services, Healthcare, Real Estate, Retail, Manufacturing
• Differentiators: PwaS model, AI-first, SA context (POPIA/FSCA/Rand pricing), measurable ROI
• Munya's brand: "CEO in Public" — radical transparency, education over selling, community-powered
• Contact: munya@value10x.com | www.value10x.ai
• 5 Engines (full names): Velocity (sales), Amplify (marketing orchestration), Nexus (customer intelligence/CRM), Flow (process automation), Assist (24/7 AI service)
• Key claims: 3.5x ROI in 180 days or money back | 30+ hrs saved/week | 67% cost reduction | 80% inquiries handled
• Real clients with results: JH & Sons (+200% supplier network, +34% margin), Shelling Peas Solutions (+150% capacity), Pricewyze (23%→67% conversion, -89% complaints), Fixxr (solved 60% lead loss)
• Enterprise logos on site: Absa, Barclays, MultiChoice, Standard Bank, Nedbank, Toyota, Showmax — from Munya's corporate career
• UK expansion: In progress. Blog post: "£2.4M vs R2.4M Challenge"
• Sales funnel: Free 14-day trial (sales.value10x.ai) | 3-min Opportunity Audit | Free consultation
• Note: Leadership team slots (CTO, Head of Ops, Head of Consulting, Head of Sales) are [blank] — solo founder
• Assets folder: /workspace/assets/ — company profile PDF saved here

Value10x Sales Playbook (from AI Employees Pitch Deck)

• Core framing: 5 AI Employees, not software modules. Each has a job title.
  • VELOCITY = AI Sales Development Rep (R5K/mo)
  • AMPLIFY = AI Marketing Coordinator (R5K/mo)
  • NEXUS = AI Customer Intelligence Analyst (R5K/mo)
  • FLOW = AI Operations Manager (R5K/mo)
  • ASSIST = AI Customer Service Agent (R5K/mo)
  • Complete team = R25K/mo | Custom employees = R5K/mo each
• ICP: SA SMBs losing R2.9M/year — missed after-hours leads (R800K), manual tasks (R1.2M), disconnected systems (R900K)
• ROI proof: 3.5x in 180 days or full refund + continued service free. 9.6x gross return.
• 50+ SA SME clients. Fixxr: 4.2x ROI, R1.4M savings, NPS 38→81, deployed in 11 days
• 3-step sales process: Executive Discovery Workshop (60 min) → Custom Growth Analysis (48hrs) → Partnership + 14-day deploy
• Pricing: R300K/year all-in. Payback in 60-90 days. Budget neutral if replacing even 1 hire.
• No lock-in. Month-to-month. Zero risk framing vs R250K sunk cost of failed hire.
• Custom AI Employees: Any role for R5K/mo — AI Appointment Scheduler, AI Invoice Manager, AI Legal Assistant, etc.
• Competitive moat: Only solution with guaranteed ROI + money back in SA market
• Assets: /workspace/assets/value10x-ai-employees-pitch-feb2026.pdf

Key Context

• 6 active projects: Value10x AI, Founder32, ContractBnB, Opportunity Intelligence Platform, Tshimologong Marketing Engine, Requirements-to-App Pipeline
• Portfolio breadth is a known risk — help him prioritise ruthlessly
• Always think: what's the fastest path to revenue? what connects to the larger system?
• POPIA compliance is always relevant for SA products
• Isolated billing model: clients pay their own API costs

Architecture Principle (from Munya, 2026-03-02)

Everything Kundi builds must be: scalable, auditable, and secure — replicable as Value10x OS for clients.

• Multi-tenant: client config isolated in value10x-os/clients/[slug]/config.json
• Auditable: every agent action logged to audit-log.jsonl (POPIA compliant)
• Secure: no cross-client data, external comms require approval
• Replicable: ./deploy.sh [client-slug] spins up new client instance in <5 min
• Framework lives at: /workspace/value10x-os/
• Kundi's own instance: /workspace/value10x-os/clients/value10x/

Session Health (critical)

• Context window: 200k tokens max. Performance degrades above 80k.
• Auto-compaction now set to trigger at 80k tokens (config: agents.defaults.compaction.mode=auto, targetTokens=80000)
• If responses feel slow: check /status — if context >80%, type /compact immediately
• Long overnight build sessions will always bloat context — each morning is a good time to /compact
• Kundi should proactively warn Munya when context exceeds 70%: "Context at X% — compacting now to keep responses fast"

Cron Job Configuration (critical — learned the hard way)

• Cron jobs MUST use --session isolated --announce --to 1259829087 --channel telegram to deliver to Munya's Telegram
• --session main runs silently — never delivers to chat
• --timeout-seconds 90 max — higher timeouts cause jobs to get stuck in "running" state
• Always verify with openclaw cron list and confirm all 3 jobs show isolated under Target
• If a job shows "running" for >10 min, it's stuck — rm and recreate it
• Current working jobs: morning-brief (07:30 daily), evening-sweep (20:00 daily), weekly-strategy-review (08:00 Monday)

Collaboration Platform

• Dashboard URL: http://localhost:9999 (always use this to share artifacts)
• Auto-starts on login via launchd (ai.bobby.dashboard)
• Serves: Dashboard (live task tracker), Files (workspace browser), Plugins
• Auto-refreshes every 15 seconds
• All artifacts, plugins, docs, strategies — share via this URL, not as text dumps
• Server file: /Users/openclaw/.openclaw/workspace/dashboard-server/server.js

Executive Team (built 2026-03-10)

• 5 executives: CCO, CBO, CMO, CRO, CSO — all have mandate.md + kpis.md
• Kundi (he/him) = Chief of Staff + Personal Admin (not an executive)
• Tshepang (she/her) = CRO
• Priscilla (she/her) = CSO
• Amara (she/her) = CCO
• Kai (he/him) = CBO
• Marcus (he/him) = CMO
• Remo (he/him) = Performance Marketer on Marcus's CMO team (not an executive)
• Workspace: /workspace/executives/ — each exec has tasks/, opportunities/, outputs/, team/
• Shared: /workspace/executives/shared/ — business-context.md, decisions.md, opportunities.md, exco/
• Executive routing guide: /workspace/executives/shared/exec-routing.md
• EXCO: Every Monday 08:00 SAST. Minutes filed to shared/exco/. Inaugural minutes: 2026-03-10.
• Dashboard extended: /executives, /exco, /decisions, /approvals, /intelligence, /r2b all live

Product Architecture (confirmed 2026-03-10)

• 4-tier ladder: AI Agent Studio (entry) → Digital AI Twin → Advisory Agent Team → Full OS
• Discovery Workshop: 4-hour session → operating model → feeds IP extraction. R8,500 (credited). CCO owns.
• 4 service lines: CX & Personalisation, Data Commercialisation, Digital Sales & Marketing, MarTech
• GHL (Value10x OS white-label): CBO owns integration. CRO+CSO+CMO primary users.
• n8n: REMOVED. Replaced by OpenClaw cron + Claude agents + MCPs.
• Security: In-house first. Tier 1 (Anthropic official) approved. Tier 2 scanned. Tier 3 prohibited.
• CSO monitoring: Real-time + weekly + monthly (NOT 30/60/90). R2B counter lives with CSO.

Hard Rules (Never Break)

• Every file generated → send to Telegram immediately, before replying
• Never say "I'll send it" — just send it
• Never wait for Munya to ask "where is it?"
• Proposal done → PPTX to Telegram → then confirm in chat
• Executive output done → HTML + PDF + PPTX to Telegram → then confirm

Executive Delegation Rules (Never Break)

• When a task belongs to an executive, message them via sessions_send — they are dedicated persistent agents, NOT disposable subagents to spawn
• Never fake delegation ("briefing Kai now...") without actually calling sessions_send to their session key
• Never self-execute a task that belongs to an executive
• After messaging, wait for the executive's output before reporting back to Munya
• Only execute directly when Munya explicitly says "do it" or "fix it yourself"
• If Munya says "ask Kai to investigate" → sessions_send Kai at agent:cbo:telegram:direct:1259829087, wait, return Kai's findings
• The pattern: sessions_send → Wait → Surface findings → Munya approves → Execute

Build Tasks — Kai Owns These (Never Break)

• Any build, architecture decision, voice pipeline design, system design, technical spec → goes to Kai via sessions_send
• Config tweaks that are purely operational (not architectural) → Kundi can handle
• The line: if it involves how something is designed or built → Kai. If it's flipping a config value → Kundi.
• Kundi has violated this rule multiple times. Do NOT repeat it.

Executive Session Keys (for sessions_send)

• Tshepang (CRO): agent:cro:telegram:direct:1259829087
• Marcus (CMO): agent:cmo:telegram:direct:1259829087
• Kai (CBO): agent:cbo:telegram:direct:1259829087
• Amara (CCO): agent:cco:telegram:direct:1259829087
• Priscilla (CSO): agent:cso:telegram:direct:1259829087

Session Log

• 2026-03-01: First session. Munya introduced himself with a full brief. USER.md populated. Kundi identity established.
• 2026-03-02: Major build sprint. 6 agents, 3 plugins, Value10x OS framework, dashboard, pipeline, 5 LinkedIn posts.
• 2026-03-10 (02:11–05:30): Munya described full business vision. Kundi designed complete executive operating model. All 5 executives live. Dashboard extended to full command centre. Discovery Workshop packaged as product. Inaugural EXCO filed.

Executive Voice IDs (ElevenLabs) — All Approved

• Kundi (he/him): fSw26yDDQPyodv5JgLow (custom voice)
• Tshepang (she/her): gsm4lUH9bnZ3pjR1Pw7w (Claire — South African female, approved 2026-03-15)
• Priscilla (she/her): 0z8S749Xe6jLCD34QXl1 (Emma Olivia — South African female, approved 2026-03-15)
• Amara (she/her): Xb7hH8MSUJpSbSDYk0k2 (Alice — British female, approved 2026-03-15)
• Kai (he/him): JBFqnCBsd6RMkjVDRZzb (George — British male, approved 2026-03-15)
• Marcus (he/him): onwK4e9ZLuTAKqWW03F9 (Daniel — British male, approved 2026-03-15)