moving

2026-09-02 10:10:29 +02:00
commit 0c3e2ead3b
3841 changed files with 970576 additions and 0 deletions
--- a/docs/wiki-live/egzakta-group.md
+++ b/docs/wiki-live/egzakta-group.md
@@ -0,0 +1,91 @@
+---
+type: entity
+entity_type: organization
+name: "Egzakta Group"
+confidence: 0.90
+sources: 30
+last_compiled: 2026-04-13T20:36:10.272Z
+frame_ids: [154, 174, 171, 167, 150, 133, 120, 92, 82, 60, 20, 787, 791, 777, 883, 41, 879, 882, 907, 861, 815, 857, 816, 814, 892, 886, 38, 870, 906, 824]
+related_entities: ["Waggle OS", "KVARK", "LM TEK", "Marko Markovic"]
+---
+
+# Egzakta Group
+
+# Egzakta Group
+
+## Summary
+
+Egzakta Group is a EUR 31M (2025 normalized) European technology holding founded in 2005 by Marko Markovic, Zoran Radisavljevic, and Nenad Tesic. The organization operates as a 4-layer sovereign AI deployment stack spanning enterprise consulting, business process management, sovereign AI orchestration, and AI infrastructure. The group is currently pursuing EUR 25M in growth equity at a EUR 100-114M pre-money valuation while scaling from ~200 to 400+ employees by 2028.
+
+## Key Facts
+
+**Corporate Structure** ([Frame #882](#frame-882))
+- Holding company: Egzakta Holding BV (Netherlands, EU-domiciled)
+- Founded: 2005 (19 years old)
+- Team size: 200+ professionals (targeting 400+ by 2028)
+- 3 founding partners: Marko Markovic, Zoran Radisavljevic, Nenad Tesic
+
+**Financial Performance** ([Frame #882](#frame-882))
+- 2025 Revenue: EUR 31M normalized
+- 2025 EBITDA: EUR 7M (23% margin)
+- 2026 Budget: EUR 46M (+49% YoY), EUR 12M EBITDA target (+70%)
+- Fundraising: Seeking EUR 25M growth equity at EUR 100-114M pre-money (18-20% dilution)
+
+**Business Architecture: 4-Layer Sovereign AI Stack** ([Frame #882](#frame-882), [Frame #907](#frame-882))
+
+1. **Egzakta Advisory** — 19-year institutional consulting practice
+   - 2026 Budget: EUR 14.5M (+113%)
+   - Pipeline: EUR 17.4M (EUR 10M+ signed)
+   - Key clients: RFZO (EUR 5.5Bn health funds), PIO (186→1 apps, 1.7M citizens), AOFI (sovereign AI platform in deployment)
+
+2. **TubeIQ** — Digital process & workflow platform
+   - 2026 Budget: EUR 7.6M
+   - 45% recurring revenue
+   - Government BPM/workflow focus, KVARK-integrated data layer
+
+3. **KVARK** — Sovereign AI orchestration platform ([Frame #787](#frame-787), [Frame #907](#frame-907))
+   - On-premises, model-agnostic, EU AI Act compliant
+   - 3 contracted clients (EUR 1.2M early 2026 revenue), 7 budgeted for 2026
+   - ACV: EUR 97-420K; 74% consolidated margin; 70% recurring
+   - Targets regulated markets (banking, utilities, government) in CEE/SEE where US cloud AI is legally prohibited
+
+4. **LM TEK / EK Water Blocks** — AI infrastructure & liquid cooling
+   - 2026 Budget: EUR 20M
+   - Historical peak: EUR 36M
+   - Acquired March 2025 (restructured)
+   - Direct-to-chip liquid cooling for GPU servers
+
+**Supporting Entities** ([Frame #882](#frame-882))
+- **RBS MBA** — Rome Business School Belgrade MBA + NEW Master in AI program (talent pipeline)
+- **eNAT** — Esports + hardware innovation lab (R&D sandbox)
+
+**Product Strategy: Waggle OS** ([Frame #41](#frame-41), [Frame #791](#frame-791), [Frame #907](#frame-907))
+- Waggle OS is the agentic frontend and demand-generation engine for KVARK
+- Intentionally free/cheap; not meant to generate direct revenue
+- Universal Memory Harvest: consolidates AI memory from 20+ platforms (ChatGPT, Claude, Gemini, etc.) into 50-100 distilled frames
+- Q2 2026 target: 50 Teams users; current blocker: Stripe production keys
+- Flywheel model: user dependency on memory → enterprise discovery → KVARK sovereign deployment contracts (EUR 400K-1.2M each)
+
+**Key Innovations** ([Frame #907](#frame-907))
+- **Universal Memory Harvest** — Multi-platform AI memory consolidation (spec v1.0 only; not yet built)
+- **EvolveSchema/GAPA** — Evolutionary schema optimization for DSPy; proven in paper (+2-4pp across 4 benchmarks); not yet integrated
+- **Sovereign AI Moat** — CEE/SEE regulatory requirement (not feature): banks, utilities, government cannot use US cloud AI; sovereign on-prem deployment mandatory
+
+**Leadership: Egzakta AI Lab** ([Frame #883](#frame-883))
+- **Marko Dadic** — Director, Head of Egzakta AI Lab; leads kvark.ai development; "AI CEO" (marko.dadic@egzakta.com)
+  - At Egzakta since July 2020 (BA → Senior BA → Consultant → Manager → Senior Manager → Director)
+- **Ivan Pakhomov** — Senior Manager, AI Lab & kvark.ai; responsible for compliance architecture and security (ivan.pakhomov@egzakta.com)
+  - Background: Deloitte, Arthur Consulting, Strelka KB; Masters in Economics from RANEPA
+
+**Key Commercial Opportunities** ([Frame #907](#frame-907))
+- EPS, Yettel, AOFI, EU Horizon 2026, Clipperton Finance
+
+## Timeline
+
+| Date | Event | Source |
+|------|-------|--------|
+| 2005 | Egzakta Group founded (19 years before 2024) | [Frame #882](#frame-882) |
+| July 2020 | Marko Dadic joins Egzakta as Business Analyst | [Frame #883](#frame-883) |
+| March 2025 | LM TEK / EK Water Blocks acquired and restructured | [Frame #882](#frame-882) |
+| Early 2026 | KVARK: 3 contracted clients, EUR 1.2M revenue | [Frame #907](#frame-907) |
+| 2026 | EUR 25M growth equity fundraising round (pre-money EUR 100-114M) | [Frame
--- a/docs/wiki-live/index.md
+++ b/docs/wiki-live/index.md
@@ -0,0 +1,36 @@
+# Wiki Index
+
+*28 pages compiled — last updated 2026-04-13*
+
+## Entities
+
+- [[Anthropic]] — 30 sources (2026-04-13)
+- [[Clerk]] — 30 sources (2026-04-13)
+- [[Data Sovereignty]] — 30 sources (2026-04-13)
+- [[EU AI Act]] — 30 sources (2026-04-13)
+- [[Egzakta Group]] — 30 sources (2026-04-13)
+- [[Fastify]] — 30 sources (2026-04-13)
+- [[Hive Mind]] — 30 sources (2026-04-13)
+- [[KVARK]] — 30 sources (2026-04-13)
+- [[LM TEK]] — 30 sources (2026-04-13)
+- [[MCP Protocol]] — 30 sources (2026-04-13)
+- [[Marko Markovic]] — 30 sources (2026-04-13)
+- [[Memory Harvest]] — 30 sources (2026-04-13)
+- [[Memory MCP]] — 30 sources (2026-04-13)
+- [[React]] — 30 sources (2026-04-13)
+- [[SQLite]] — 30 sources (2026-04-13)
+- [[Tauri]] — 30 sources (2026-04-13)
+- [[Tier Strategy]] — 30 sources (2026-04-13)
+- [[TypeScript]] — 30 sources (2026-04-13)
+- [[Waggle OS]] — 30 sources (2026-04-13)
+- [[Wiki Compiler]] — 30 sources (2026-04-13)
+
+## Cross-Source Synthesis
+
+- [[Synthesis: Data Sovereignty]] — 30 sources (2026-04-13)
+- [[Synthesis: EU AI Act]] — 30 sources (2026-04-13)
+- [[Synthesis: KVARK]] — 30 sources (2026-04-13)
+- [[Synthesis: Memory Harvest]] — 30 sources (2026-04-13)
+- [[Synthesis: Tier Strategy]] — 30 sources (2026-04-13)
+- [[Synthesis: Waggle OS]] — 30 sources (2026-04-13)
+- [[Synthesis: Wiki Compiler]] — 30 sources (2026-04-13)
--- a/docs/wiki-live/kvark.md
+++ b/docs/wiki-live/kvark.md
@@ -0,0 +1,78 @@
+---
+type: entity
+entity_type: project
+name: "KVARK"
+confidence: 0.90
+sources: 30
+last_compiled: 2026-04-13T20:35:51.259Z
+frame_ids: [21, 792, 788, 789, 784, 787, 836, 791, 790, 907, 793, 883, 930, 31, 879, 856, 849, 826, 46, 1005, 851, 844, 174, 923, 23, 171, 882, 37, 848, 29]
+related_entities: ["Data Sovereignty", "Egzakta Group", "Waggle OS", "Marko Markovic"]
+---
+
+# KVARK
+
+# KVARK
+
+## Summary
+
+KVARK is Egzakta Group's sovereign enterprise AI platform, deployed on-premises or in private clouds for regulated markets that cannot use US-based cloud AI. It serves as the enterprise monetization endpoint for Waggle OS, a freemium demand-generation SaaS product with persistent memory capabilities. As of early 2026, KVARK has EUR 1.2M in contracted revenue across 3–4 early-stage clients and represents the core of Egzakta's EUR 46M revenue budget for 2026.
+
+## Key Facts
+
+**Product & Architecture**
+- Everything Waggle OS does, deployed on your infrastructure with full data governance and audit trails ([Frame #787](source://frame/787))
+- Fully on-premises AI without cloud dependencies; enables sovereign deployment required by CEE/SEE regulated markets (banking, utilities, government) ([Frame #792](source://frame/792), [Frame #907](source://frame/907))
+- Full Microsoft 365 integration, EU AI Act compliance by default, custom model pools, complete audit trail ([Frame #790](source://frame/790))
+- Tech stack: Qwen inference (vLLM), LiteLLM proxy, Qdrant vector database, enterprise connectors, BPMN orchestration ([Frame #907](source://frame/907))
+
+**Revenue & Commercials**
+- EUR 1.2M in contracted revenue as of April 2026 ([Frame #788](source://frame/788))
+- Business model: Waggle OS is freemium SaaS (demand generation) → KVARK is enterprise consultative sale ([Frame #793](source://frame/793))
+- ACV EUR 97–420K; 74% consolidated margin; 70% recurring revenue ([Frame #882](source://frame/882))
+- 4 paying clients as of early 2026, 7 budgeted for 2026 ([Frame #882](source://frame/882))
+
+**Hardware & Infrastructure**
+- LM TEK is the hardware arm providing GPU infrastructure (liquid-cooled servers, EK Fluid Works brand) ([Frame #792](source://frame/792))
+- Boston Limited channel partner for NVIDIA relationships ([Frame #907](source://frame/907))
+
+**Waggle → KVARK Flywheel**
+- Waggle harvests AI memory from 20+ platforms for free, creating user dependency ([Frame #907](source://frame/907))
+- Users learn AI-native work in Waggle, then enterprises want it on their infrastructure ([Frame #788](source://frame/788))
+- Teams tier ($49/seat/mo) includes KVARK connector nudges; Enterprise tier converts to consultative KVARK sales ([Frame #930](source://frame/930), [Frame #784](source://frame/784))
+- Memory + Harvest are free forever in Waggle—the lock-in moat that drives KVARK pipeline ([Frame #793](source://frame/793))
+
+**Launch Status (as of 2026-03-23)**
+- V1 launch requires: code signing (SSL.com eSigner EV or Certum SimplySign), KVARK HTTP API wiring (3 core endpoints: login, me, search) ([Frame #836](source://frame/836))
+- KvarkClient library exists with 30 mocked tests; real API connection not yet wired to production KVARK backend ([Frame #836](source://frame/836), [Frame #849](source://frame/849))
+- Settings panel UI exists for KVARK configuration; production wiring deferred to enterprise phase ([Frame #849](source://frame/849))
+
+**Strategic Context**
+- Part of Egzakta Group's 4-layer sovereign AI deployment stack (Advisory → TubeIQ → KVARK → LM TEK infrastructure) ([Frame #882](source://frame/882))
+- Regulatory moat: CEE/SEE regulated markets (banking, utilities, government) legally cannot use US cloud AI; sovereign deployment is a regulatory requirement, not a feature ([Frame #907](source://frame/907))
+- Key opportunities: EPS, Yettel, AOFI, EU Horizon 2026, Clipperton Finance ([Frame #907](source://frame/907))
+
+## Timeline
+
+| Date | Event | Source |
+|------|-------|--------|
+| Early 2026 | KVARK reaches EUR 1.2M contracted revenue (3–4 clients) | [Frame #788](source://frame/788) |
+| 2026-03-20 | V1 Launch Backlog finalized; product ship-ready for internal testing | [Frame #836](source://frame/836) |
+| 2026-04-12 | Tier restructure confirmed (Trial/Free/Pro/Teams/Enterprise) | [Frame #930](source://frame/930) |
+| 2026 | Target: 7 KVARK clients (budgeted), EUR 46M group revenue | [Frame #882](source://frame/882) |
+
+## Relations
+
+- **Egzakta Group** — parent company, founded by Marko Markovic ([Frame #791](source://frame/791), [Frame #882](source://frame/882))
+- **Waggle OS** — freemium demand-generation frontend; feeds KVARK sales pipeline ([Frame #788](source://frame/788), [Frame #907](source://frame/907))
+- **LM TEK** — hardware sister company providing GPU infrastructure ([Frame #792](source://frame/792))
+- **Marko Markovic** — founder/overall strategy + enterprise sales ([Frame #907](source://frame/907))
+- **Marko Dadic** — Director, Head of Egzakta AI Lab; leads kvark.ai development ([Frame #883](source://frame/883))
+- **Ivan Pakhomov** — Senior Manager, AI Lab; responsible for compliance architecture and security ([Frame #883](source://frame/883))
+- **EU AI Act compliance** — key product differentiator for regulated markets ([Frame #790](source://frame/790))
+- **Microsoft 365 integration** — core enterprise connector capability ([Frame #790](source://frame/790))
+
+## Open Questions
+
+1. **Real KVARK backend status**: API requirements are documented ([Frame #836](source://frame/836)), but what is the actual deployment status of KVARK's HTTP API endpoints? When will production KVARK backend accept Waggle client connections?
+
+2. **Paid clients vs. contracted
--- a/docs/wiki-live/marko-markovic.md
+++ b/docs/wiki-live/marko-markovic.md
@@ -0,0 +1,81 @@
+---
+type: entity
+entity_type: person
+name: "Marko Markovic"
+confidence: 0.90
+sources: 30
+last_compiled: 2026-04-13T20:35:14.293Z
+frame_ids: [174, 171, 167, 811, 812, 20, 791, 826, 1013, 854, 892, 883, 886, 810, 876, 809, 901, 868, 29, 882, 873, 42, 847, 902, 43, 907, 44, 1001, 16, 891]
+related_entities: ["Egzakta Group", "Waggle OS", "KVARK"]
+---
+
+# Marko Markovic
+
+# Marko Markovic
+
+## Summary
+
+Marko Markovic is a strategy consultant and entrepreneur with 25+ years of experience in banking, telecom, and government sectors. He is a founding partner of Egzakta Group, a EUR 31M technology holding company, and leads development of Waggle OS and KVARK, a sovereign AI orchestration platform. Known for direct, data-driven communication and production-readiness standards, Marko operates across enterprise consulting, product development, and thought leadership on AI governance and sovereignty.
+
+## Key Facts
+
+**Professional Background**
+- Strategy consultant and AI strategist with 25+ years experience in regulated markets (banking, telecom, government) [Frame #826]
+- Founding partner at Egzakta Advisory, part of Egzakta Group holding company [Frame #854, #882]
+- Title: Consultant, entrepreneur, AI strategist, solution provider [Frame #826]
+- Email: marolinik@gmail.com [Frame #826]
+- Based in Serbia, operates in CEE/SEE region with European expansion focus [Frame #882]
+
+**Current Projects & Leadership**
+- Leads Waggle OS development — a universal memory harvesting platform targeting Teams integration with 50-user milestone by end of Q2 2026 [Frame #174, #881]
+- Leads KVARK development — sovereign AI orchestration platform (on-premise, model-agnostic, EU AI Act compliant) with 4 contracted clients generating EUR 1.2M early 2026 revenue [Frame #907, #882]
+- Founder of Egzakta Group (2005, 19 years established) with co-founders Zoran Radisavljevic and Nenad Tesic [Frame #882]
+- Egzakta Group: 200+ employees, EUR 31M normalized revenue (2025), EUR 46M budget (2026) [Frame #882]
+- Seeking EUR 25M growth equity at EUR 100-114M pre-money valuation [Frame #882]
+
+**Technical Skills & Preferences**
+- Prefers TypeScript over JavaScript; works with Node.js backend systems [Frame #174]
+- Expert-level understanding of architecture, databases, API design, React, TypeScript [Frame #854]
+- Prefers dark mode interfaces; team uses Slack for communication [Frame #174]
+- Values real integration tests over mocked tests; learned from failed production migrations [Frame #811]
+- Prefers bundled PRs over many small ones for refactoring work [Frame #812]
+
+**Communication & Work Style**
+- Direct, no-fluff, data-driven communication preference [Frame #854]
+- Prefers bullet-point executive summaries over long prose [Frame #873]
+- Operates at production-readiness level: no "for now" shortcuts, everything must be bulletproof [Frame #854, #1013]
+- Terse, action-oriented; no process summaries—prefers to read diffs [Frame #809]
+- When he says "you decide" or "you lead," takes initiative without asking permission [Frame #810]
+- Regularly requests 20-30 item sprints in single sessions and expects all completed [Frame #847]
+- Uses ULTRATHINK flag for complex tasks expecting deep analysis before coding [Frame #854]
+- Comfortable with large scope; expects granular, well-described commits [Frame #854]
+- Demands explicit confirmation for push to master [Frame #854]
+
+**Testing & Quality Standards**
+- Zero tolerance for regressions; all 4,333 tests must pass [Frame #854]
+- Uses structured UAT rounds (R1/R2/R3) to validate work systematically [Frame #854]
+- Leads via Playwright E2E testing; prefers automated test execution over manual click-through verification [Frame #902]
+
+**Content & Thought Leadership**
+- Building personal brand as opinion maker/influencer with high-traction social media presence [Frame #826]
+- Favorite movie: The Matrix; views Agent Smith as representing corporate consulting gone wrong—interesting metaphor given his consulting work [Frame #873]
+- 106 sources tagged across 8 thesis pillars in NotebookLM "Sovereign AI Thesis" notebook (orchestration, enterprise, sovereignty, infrastructure, governance, geopolitics, economics, voice) [Frame #901]
+- LinkedIn engagement tracker: 32 → 37 as of Apr 12 session [Frame #876]
+- Uses Mixpost for LinkedIn scheduling (36+ posts queued) and Blotato for YouTube, Instagram, TikTok, company pages [Frame #886]
+- Produces newsletters grounded in NotebookLM citations (Edition #2: "The Board Approved the Budget, Not the Governance" ~1,550 words, queued Wed Apr 15) [Frame #876]
+
+**Platform & Infrastructure**
+- Domain: mysocial.my; hosting via Hostinger VPS (Frankfurt) [Frame #826]
+- Microsoft Graph API access via Claude-Graph-Bridge Azure AD app (tenant: 151b2032..., client: 7bac5741...) authenticated as marko.markovic@egzakta.com [Frame #892]
+- LinkedIn accounts: Primary profile (Marko Markovic, MSc, MBA, ID 17884) with 5 company subaccounts [Frame #886]
+- YouTube: Marko Markovic (ID 33380); Instagram: markovic2324 (ID 40995); TikTok: marolinik74 (ID 37963); X: MarkoMarko36764 (ID 16203) [Frame #886]
+- Blotato MCP connected to 5 platforms with 14 tools and visual templates (Alan Ford comics, infographics, video templates) [Frame #886]
+
+**Team & Collaboration**
+- Works with Marko Dadic (Director, Head of Egzakta AI Lab, leads KVARK) and Ivan Pakhomov (Senior Manager, responsible for compliance/security in KVARK) [Frame #883]
+- Standard pattern: CC both Marko Dadic and Ivan Pakhomov on all partnership, BD, and technical collaboration emails [Frame #883]
+- Team lead for isolation project: Ana Petrović [Frame #873]
+- Collaborates with Mihail on GAPA + KVARK backend and EvolveSchema paper [Frame #907]
+
+**Strategic Context**
+- Egzakta Group operates 4-layer sovereign AI deployment stack
--- a/docs/wiki-live/memory-harvest.md
+++ b/docs/wiki-live/memory-harvest.md
@@ -0,0 +1,114 @@
+---
+type: concept
+name: "Memory Harvest"
+confidence: 0.85
+sources: 30
+last_compiled: 2026-04-13T20:36:35.605Z
+frame_ids: [34, 22, 24, 21, 784, 793, 905, 808, 918, 1005, 799, 930, 915, 925, 907, 174, 917, 924, 171, 927, 786, 167, 803, 796, 926, 797, 154, 928, 781, 150]
+related_entities: ["Memory Harvest"]
+---
+
+# Memory Harvest
+
+# Memory Harvest
+
+## TL;DR
+
+Memory Harvest is Waggle OS's strategic moat—a universal import system that consolidates AI conversations from 20+ platforms into a single governed workspace. It's free forever and drives adoption by solving the fragmentation problem enterprises face with scattered AI interactions.
+
+## What We Know
+
+### Core Definition & Purpose
+
+Memory Harvest is a four-pass LLM distillation pipeline that imports conversations from external AI systems (ChatGPT, Claude, Gemini, etc.) and converts them into Waggle's 5-layer memory format (I/P/B frames + Knowledge Graph + compliance metadata) [[#917, #905, #793]].
+
+**Strategic role:** It's the primary demand-generation hook for KVARK. Enterprises adopt Waggle for memory consolidation, discover it solves EU AI Act compliance (audit trails, Art. 12/14/19/26/50 tracking), then graduate to sovereign KVARK deployment [[#793, #1005, #907]]. This compounds into the TAM expansion: Waggle is free moat → Teams is $49/seat/mo → KVARK is EUR 400K–1.2M per enterprise [[#930, #793]].
+
+### Architecture & Implementation
+
+**Status:** Fully implemented and tested as of 2026-04-10 [[#905]].
+
+**Pipeline layers:**
+1. **5 source adapters** (Phase 1 complete):
+   - ChatGPT JSON export
+   - Claude conversation export
+   - Claude Code filesystem ingestion
+   - Gemini takeout
+   - Auto-detect universal format
+   
+   Planned Phase 2: 30+ additional adapters across 3 tiers (markdown, plaintext, PDF, URL, plus Microsoft Graph, Slack, Notion, Gmail, GitHub via MCP) [[#803, #928]].
+
+2. **4-pass distillation:**
+   - Pass 1: Raw extraction into Universal Import Format (UIF)
+   - Pass 2: Deduplication & chunking
+   - Pass 3: Entity extraction + Knowledge Graph construction
+   - Pass 4: Compliance metadata tagging [[#917]]
+
+3. **Storage & retrieval:**
+   - All frames persisted as I-frames (Immutable) with `gop_id = 'harvest'`
+   - Vector embeddings (sqlite-vec, 1024 dims, Xenova/all-MiniLM-L6-v2)
+   - FTS5 keyword search + semantic search
+   - HarvestSourceStore tracks source origin, sync timestamp, auto-sync toggle [[#905, #925]]
+
+### Performance & Reality Check
+
+**E2E verified (as of 2026-04-10):**
+- 156 frames extracted in 18 ms, persisted in 76 ms
+- Idempotent re-runs confirmed (second run = 0 inserts)
+- 10 real FTS5 queries validated against harvest frames
+- Vector semantic search tested ("my identity and name" → User Profile, "where did we deploy mixpost" → 3 Mixpost memories)
+- 81/81 E2E tests passing (phase-ab + full-product-audit + power-user-stress) [[#905, #925]]
+
+**Data cleanup critical:** Wiki compilation test (2026-04-13) found personal.mind was 80% E2E test pollution, reducing usable signal. Real user data needed for >80% hit rate [[#928]].
+
+### Monetization & Lock-In
+
+Memory Harvest is **free forever** — not a paid feature [[#784, #799, #793]].
+
+**Why:** It's the lock-in moat. Users become dependent on unified memory → switching costs rise → Teams upsell ($49/seat/mo for shared team memory) becomes natural → enterprise compliance needs trigger KVARK path.
+
+Free forever decision removes per-tier quotas on embedding capacity—all tiers get unlimited [[#799]].
+
+### Compliance Integration
+
+Harvest pipeline automatically tags frames with EU AI Act metadata:
+- Article 12 (transparency): logged in compliance table
+- Article 14 (accuracy): source + extraction method tracked
+- Article 19 (human oversight): agent decision points marked
+- Articles 26/50 (rights + remedies): audit trail immutable [[#917, #1005]]
+
+ComplianceDashboard shows risk classification per workspace [[#917]].
+
+### Known Gaps & Future Work
+
+**Memory MCP Plugin (5 fixes completed 2026-04-13):**
+- Workspace mind layer caching with LRU invalidation
+- `scope='all'` now searches ALL workspaces (not just one)
+- Default to mock embeddings (Ollama > API keys > mock) to avoid 23MB surprise downloads
+- Reliable dedup via batch-start timestamp
+- Daily sessions instead of eternal sessions [[#928]]
+
+**Remaining P1 priorities:**
+- Shared Team Memory (teams-tier killer feature)
+- Cross-workspace read permissions & approval modal
+- Post-harvest auto-run cognify pipeline on imports [[#915, #927]]
+
+**Phase 2 (scaling):** 30+ adapters, Microsoft Graph full stack, webhook/file-watch/cron automations [[#928]].
+
+## Sources & Evolution
+
+### Initial Spec → Implementation
+
+Frame #1005 (2026-04-13) documented Memory Harvest as **spec-only**, the strategic weapon needed to launch Waggle. Frame #917 (2026-04-10) proved it works end-to-end on real data—156 frames extracted in 18 ms with zero production bugs. Frame #905 validated the full pipeline: extraction, persistence, FTS5, vector search, Knowledge Graph construction.
+
+### Monetization Crystallization
+
+Early frames discussed per-tier embedding quotas [[#799]], but 2026-04-12 decision unified the tier strategy: Memory + Harvest = free forever, agents = free, marketplace skills/connectors = paywall, Teams = shared memory ($49/seat), Enterprise = KVARK [[#930]]. This clarified why Harvest is moat—it drives adoption without revenue cannibalization.
+
+### Wiki Compilation Reality Check
+
+2026-04-13 wiki compilation test (Frame #928) revealed that raw harvested data alone doesn't generate 80%+ useful wiki pages. Cross-session synthesis works, but data quality matters. This suggests future work should prioritize data cleaning on import (dedup, time-series aggregation, PII filtering).
+
+## Related Topics
+
+- [[Waggle
--- a/docs/wiki-live/synthesis-waggle-os.md
+++ b/docs/wiki-live/synthesis-waggle-os.md
@@ -0,0 +1,94 @@
+---
+type: synthesis
+name: "Synthesis: Waggle OS"
+confidence: 0.65
+sources: 30
+last_compiled: 2026-04-13T20:37:17.738Z
+frame_ids: [154, 174, 171, 167, 150, 133, 120, 92, 39, 82, 41, 60, 796, 780, 791, 788, 787, 786, 793, 789, 801, 805, 832, 846, 907, 37, 38, 844, 836, 845]
+related_entities: ["user_stated", "import"]
+---
+
+# Synthesis: Waggle OS
+
+# Waggle OS — Cross-Source Synthesis
+
+## Cross-Source Summary
+
+**Waggle OS** is a workspace-native AI agent platform with persistent memory, built by Egzakta Group (founded by Marko Markovic) in Serbia. It ships as a Tauri 2.0 desktop binary (Windows/macOS) with a React frontend and Node.js sidecar. The product is currently in internal testing (friends & family) post-Milestone 2, with a freemium SaaS model targeting solo users first, then teams, with an enterprise upsell to KVARK (Egzakta's sovereign enterprise AI platform).
+
+Core technical facts across all sources:
+- **Architecture**: Tauri desktop app + Node.js backend, React UI, persistent SQLite `.mind` files with FTS5 + sqlite-vec
+- **Release Status**: M2 complete (232 tests, 7 commits), 10.6 MB exe, production audit passed (57 findings fixed, 9.7/10 confidence)
+- **Team**: TypeScript/Node.js stack preference, uses Slack for comms, targets Teams/Slack integrations
+- **Business**: Freemium tier funnel (Free → $15/mo → $79/mo Teams), KVARK enterprise consultative sale (EUR 1.2M contracted revenue)
+- **Launch Blocker**: Code signing (SSL.com EV recommended) before public release
+
+## Patterns (Recurring Themes)
+
+### 1. **Memory-First Architecture**
+Every source emphasizes persistent memory as the core differentiator:
+- Personal mind + workspace mind + collective mind (frames #780, #846, #907)
+- FTS5 + sqlite-vec + knowledge graph (frames #832, #846)
+- "Memory Harvest" monetization moat — universal import from 20+ platforms (frames #793, #788, #801, #907)
+
+### 2. **Waggle Dance Protocol (Agent-to-Agent Communication)**
+Consistent across multiple sources:
+- Bee metaphor as actual architecture (frame #846)
+- Agents share discoveries via structured protocol, not just chat (frames #846, #856 inferred)
+- Knowledge flows: personal → workspace → collective → team (frame #846)
+
+### 3. **Two-Product Flywheel**
+Strategic alignment across all recent frames (#787-#907):
+- Waggle = demand-gen (freemium, locks users in with memory)
+- KVARK = enterprise play (sovereign deployment, EUR 400K-1.2M contracts)
+- Users learn on Waggle → enterprises adopt KVARK (frame #793, #788, #907)
+
+### 4. **Regulatory/Sovereign Moat**
+CEE/SEE markets (banking, utilities, government) cannot use US cloud AI legally. This is business context, not feature (#907).
+
+### 5. **Velocity Through Isolation Testing**
+Early frames show aggressive benchmark/isolation testing (frames #154-#174, repeated bench-secrets). This appears to be stress-testing memory isolation across sessions, not actual user data.
+
+## Contradictions
+
+**None detected within substantive claims.** 
+
+However, there is a **temporal inconsistency**:
+- Early frames (#154, #174, etc.) dated 2026-04-04 to 2026-04-05 show repetitive benchmark data with "0 corrections" and "0 interactions"
+- Later frames (#832+) dated 2026-04-13 show completed M2 with 232 tests, shipped v1.0.0 npm package, 57 audit findings fixed
+- **Interpretation**: Early frames appear to be automated/synthetic test data (isolation & integrity tests, repeated confidential markers). Late frames are substantive project updates. No contradiction in product claims, just data quality layers.
+
+**Minor nuance** (not contradiction):
+- Frame #793 says Waggle is "intentionally free/cheap. Not meant to generate direct revenue."
+- Frame #788 says Waggle has "50 Teams users target by end of Q2" with Stripe prod keys blocker
+- **Resolution**: Both true — Waggle is low-margin demand-gen; Teams tier exists for institutional lock-in, not primary revenue source.
+
+## Insights
+
+### 1. **The Architecture Implies Offline-First Sync**
+All frames describe `.mind` files as local SQLite stores, but Waggle Dance protocol and team minds require multi-user sync. Frames don't explicitly mention CRDT/OT or sync strategy, but the architecture implies eventual consistency via the Waggle Dance protocol.
+
+### 2. **Production Hardening Completed, But Distribution Incomplete**
+Frame #844 shows 57 audit findings fixed (9.7/10 confidence), npm published, but frame #836 shows **code signing is still the launch blocker**. This suggests product-market readiness is ready, but regulatory/trust hurdles (SmartScreen, EU CA validation) are the gate.
+
+### 3. **Memory Harvest is the Intellectual Moat, Not the Technology**
+Frames #801 and #907 compare Waggle's Wiki Compiler to competitors (Mem0, Zep, Hindsight, Cognee), but the real moat is:
+- **Universal harvest** (20+ platform import)
+- **Source provenance** (tracking where knowledge came from)
+- **Schema evolution** (GAPA+EvolveSchema paper)
+
+The actual memory *storage* (FTS5 + sqlite-vec) is commodity. Competitors have similar. The differentiation is in *curation* and *evolutionary optimization*.
+
+### 4. **Waggle Dance Protocol is Central, But Not Yet Documented**
+Frames describe it metaphorically (#846), but no specification found. This is likely a critical blocker for:
+- Team mode (M3) which requires inter-agent messaging
+- Open-source contribution (external agents need to speak the protocol)
+- KVARK integration (sovereign nodes need to federate)
+
+### 5. **KVARK Wiring is Mostly Mocked**
+Frame #836: "KvarkClient library code exists with 30 mocked tests" + "HTTP API requirements delivered" but "not integrated." This means:
+- M2 shipped without KVARK connectivity
+- Enterprise tier is architecturally ready but functionally stub
+- M3 (Team Pilot) is likely the real KVARK integration point
+
+### 6. **The "Test User" Frames are Isolation/Benchmark Data**
--- a/docs/wiki-live/waggle-os.md
+++ b/docs/wiki-live/waggle-os.md
@@ -0,0 +1,81 @@
+---
+type: entity
+entity_type: project
+name: "Waggle OS"
+confidence: 0.90
+sources: 30
+last_compiled: 2026-04-13T20:35:34.634Z
+frame_ids: [154, 174, 171, 167, 150, 133, 120, 92, 39, 82, 41, 60, 832, 796, 846, 907, 37, 780, 38, 844, 791, 836, 788, 845, 787, 786, 848, 852, 789, 1005]
+related_entities: ["EU AI Act", "KVARK", "Tier Strategy", "Memory MCP", "Wiki Compiler", "Memory Harvest", "React", "TypeScript", "Tauri", "SQLite", "Fastify", "Clerk", "Egzakta Group", "Marko Markovic"]
+---
+
+# Waggle OS
+
+# Waggle OS
+
+## Summary
+
+**Waggle OS** is a desktop-native AI agent platform with persistent structured memory, built by Egzakta Group as the demand-generation engine for KVARK (the company's sovereign enterprise AI platform). Shipping as a Tauri 2.0 desktop application with React frontend and Node.js sidecar, Waggle enables individuals and teams to manage AI conversations across 20+ platforms through unified memory, audit trails, and agent-native workflows. The product is technologically mature (4,409 passing tests, zero TypeScript errors as of March 2026) but requires UI/UX refinement and Memory Harvest feature completion before public launch.
+
+---
+
+## Key Facts
+
+**Technical Foundation**
+- Built with Tauri 2.0 (Rust desktop + React frontend + Node.js sidecar) [Frame #780]
+- Language preference: TypeScript for development [Frame #154]
+- Persists memory in SQLite `.mind` files with FTS5 + sqlite-vec + knowledge graph [Frame #846]
+- Includes Memory MCP plugin (18 tools, 4 resources) compatible with Claude Code, Claude Desktop, and any MCP client [Frame #786]
+- Test suite: 4,409 passing tests, 100% pass rate as of March 2026; 4,185+ tests by Wave 11 [Frame #844, #848]
+- TypeScript errors: 0 (fixed from 87 in production hardening wave) [Frame #844]
+
+**Product Architecture**
+- Five-layer memory system: personal.mind, workspace.mind, collective.mind, team minds [Frame #846]
+- Agents communicate via **Waggle Dance** protocol for inter-agent memory and knowledge exchange [Frame #846]
+- 80+ tools, 22 personas, skill SDK with manifest system [Frame #844, #845]
+- Model router supporting multi-provider LLM access [Frame #832]
+- Interactive CLI (REPL, slash commands, markdown rendering) [Frame #832]
+- Plugin system with manifest validation, install/uninstall [Frame #832]
+
+**Project Status (as of March 2026)**
+- **Milestone 0**: Complete (172 tests) [Frame #832]
+- **Milestone 1**: Complete — desktop app shipping as 10.6 MB waggle.exe binary [Frame #832]
+- **Milestone 2**: Complete — 232 tests, 7 commits, 6 npm packages (@waggle/core, agent, optimizer, weaver, cli, sdk) [Frame #832]
+- **Gap Register**: 68 of 69 items implemented in March 2026; 3 items deferred to enterprise phase (KVARK integration, SSO, Tauri build verification) [Frame #848]
+- **Wave 11 Production Hardening** (2026-03-20): 57 audit findings fixed (8 CRITICAL → 0); security, agent rate-limiting, frontend error boundaries, accessibility hardened [Frame #844]
+- **npm Publishing**: `@waggle-ai/waggle` v1.0.0 published on npmjs.com [Frame #844]
+
+**Launch Readiness**
+- Internal testing phase (friends & family) is active; product ready for unsigned distribution [Frame #836]
+- **Blocking public launch**: Code signing (recommendation: SSL.com eSigner EV ~$240/yr or Certum SimplySign EV ~$249/yr using Netherlands entity) [Frame #836]
+- KVARK wiring specified (6 HTTP endpoints, 3 P0 for MVP: login, me, search) — awaiting KVARK API availability [Frame #836]
+- All deferred items non-blocking for internal testing [Frame #836]
+
+**UI/UX Status**
+- Two CSS systems diverge on design tokens; 360+ instances of 9-10px text reduce readability [Frame #1005]
+- Stripe billing integration has 5 blocking gaps [Frame #1005]
+- Memory Harvest (consolidate conversations from 20+ platforms) still spec-only, not built [Frame #1005]
+- SplashScreen on-brand, light theme fixed, all hardcoded hex values → theme tokens in Wave 11 [Frame #844]
+- Accessibility (ARIA) hardened; App.tsx refactored (1338 → 1140 lines, 6 hooks extracted) [Frame #844]
+
+**Business Model & Revenue**
+- **Tier funnel**: Solo (Free) → Basic ($15/mo) → Teams ($79/mo) → Enterprise (consultative KVARK sale) [Frame #789]
+- KVARK context: EUR 1.2M in contracted revenue; Waggle is intentionally free/cheap, not direct revenue driver [Frame #788, #846]
+- Egzakta Group: ~200 employees, EUR 4.5M EBITDA, targeting EUR 10M in 2026 [Frame #846]
+- Key enterprise opportunities: EPS, Yettel, AOFI, EU Horizon 2026, Clipperton Finance [Frame #846]
+
+**Strategic Vision**
+- Two core hooks for market penetration:
+  1. **"Bring Your Memory Home"** — Memory Harvest consolidates 20+ platform conversations into governed workspace [Frame #1005]
+  2. **"AI Act Compliance by Default"** — All work auditable inside Waggle; no bolt-on governance needed [Frame #1005]
+- Sovereign deployment moat: CEE/SEE regulated markets (banking, utilities, government) legally cannot use US cloud AI; Waggle + KVARK meet regulatory requirement [Frame #846]
+- Universal Memory Harvest spec v1.0 complete; implementation planned for post-launch phase [Frame #846]
+- EvolveSchema/GAPA evolutionary schema optimization proven in research (+2-4 pp across benchmarks); research phase complete, integration pending [Frame #846]
+
+**Team & Governance**
+- Built by Egzakta Group (founded by Marko Markovic) [Frame #791]
+- Key people: Marko (strategy + enterprise sales), Mihail (GAPA + KVARK backend), Marko D. (Waggle architecture), Ivan (LM TEK + hardware) [Frame #846]
+- CLI/config system supports `~/.waggle/` home directory [Frame #832]
+- Slack for team communication [Frame #38]
+
+**Deferred/
--- a/docs/wiki-live/wiki-compiler.md
+++ b/docs/wiki-live/wiki-compiler.md
@@ -0,0 +1,119 @@
+---
+type: concept
+name: "Wiki Compiler"
+confidence: 0.85
+sources: 30
+last_compiled: 2026-04-13T20:36:56.109Z
+frame_ids: [797, 800, 801, 802, 932, 928, 929, 808, 960, 965, 986, 36, 945, 950, 905, 944, 22, 990, 21, 3, 24, 174, 31, 4, 171, 45, 1, 28, 167, 42]
+related_entities: ["Wiki Compiler"]
+---
+
+# Wiki Compiler
+
+# Wiki Compiler
+
+## TL;DR
+
+Wiki Compiler is a Karpathy-style LLM system that synthesizes accumulated memory frames into persistent, interlinked markdown wiki pages. It combines structured memory storage (FrameStore + knowledge graph) with automatic compilation into human-readable entity, concept, and synthesis pages—the missing "READ" side of personal knowledge management.
+
+## What We Know
+
+### Core Concept & Inspiration
+
+Wiki Compiler draws from **Karpathy's LLM Wiki pattern** (April 2026 decision, Frame #797), which uses LLMs to incrementally build a persistent markdown wiki from raw knowledge. The key insight: answers should compound into permanent reference material rather than disappearing into chat history. This is paired with **Waggle's memory engine** (FrameStore + hybrid search + knowledge graph), creating the first complete read-write cycle for a personal knowledge OS (Frame #932).
+
+### Architecture & Compilation
+
+The system uses a **6-layer architecture** (Frame #932):
+1. **Raw sources** → ingest from markdown, PDF, URLs, plaintext, code, emails, Slack, Notion
+2. **FrameStore** → structured memory frames with timestamps and types
+3. **Knowledge Graph** → typed entities with relationships
+4. **HybridSearch** → keyword + semantic retrieval across all sources
+5. **Compilation engine** → LLM synthesis with citations
+6. **Wiki output** → entity/concept/synthesis/index/health pages in markdown
+
+### Page Types
+
+Wiki Compiler generates **five distinct page types** (Frame #802):
+- **Entity pages** — person, project, organization (e.g., "Waggle," "Karpathy")
+- **Concept pages** — topic synthesis (e.g., "EU AI Act Compliance," "Memory MCP")
+- **Synthesis pages** — cross-source patterns (the "killer feature," Frame #800) detecting recurring themes across multiple sources
+- **Index pages** — navigable catalogs with summaries
+- **Health pages** — contradictions, gaps, orphans, data quality scores
+
+### Universal Source Pipeline
+
+Frame #801 identifies a **multi-tier ingest strategy**:
+- **Tier 1** (core): Markdown, PDF, URL, plaintext
+- **Tier 2** (expanded): Obsidian vault imports, code artifacts, API exports
+- **Tier 3** (live feeds): Email, Slack, Notion, GitHub via 148+ MCP connectors
+
+The system auto-detects source type and routes to appropriate adapter (MarkdownAdapter, PlaintextAdapter, UrlAdapter, PdfAdapter, ClaudeCodeAdapter per Frame #929).
+
+### Regulatory Compliance Layer
+
+Frame #932 highlights **EU AI Act compliance** as a differentiator. The second brain itself serves as an audit trail:
+- Articles 12/13/14/19/26/50 mapped to system components
+- Auto-generated compliance pages
+- Regulatory export packages for legal review
+- PII filtering and GDPR-compliant data retention
+
+### Implementation Status (v1 BUILD Complete)
+
+As of 2026-04-13 (Frames #929, #932):
+
+**Shipped in v1 BUILD:**
+- `@waggle/wiki-compiler` package (941 lines, Frame #929)
+- 5 core compilation methods: `compileEntityPage()`, `compileConceptPage()`, `compileSynthesisPage()`, `compileIndex()`, `compileHealth()`
+- 4 source adapters: Markdown, Plaintext, URL, PDF
+- ClaudeCodeAdapter with decision extraction (12 regex patterns)
+- 2 cleanup MCP tools: `cleanup_frames`, `cleanup_entities`
+- 1 ingest MCP tool: `ingest_source` (auto-detection)
+- `CompilationState` with SQLite watermarks + incremental compilation
+- Live test on 29 memory files + 422 messages: **57% hit rate** (4/7 pages genuinely useful, Frame #928)
+
+**Architecture highlights:**
+- Incremental compilation via watermarks (no re-processing entire corpus)
+- Cross-ref linker + linter (detects orphan entities, contradictions)
+- MCP tool integration (compile/search/lint available to agents)
+- Privacy-first: local embeddings, optional Ollama/API fallback, zero telemetry
+
+### Competitive Position
+
+Frame #801 analyzed the landscape: **Google Brain** (markdown+PGLite but no real code), **Mem0** (facts graph), **Zep** (temporal), **Hindsight** (auto-capture), **Cognee** (scientific). Wiki Compiler combines:
+- Hybrid search + typed knowledge graph
+- Multi-workspace + team sync capability
+- Universal harvest pipeline (30+ adapters)
+- Compiled wiki output
+- **AI Act compliance mapping** (unique)
+
+No competitor combines all six elements (Frame #801).
+
+### Dual-Track Delivery
+
+- **Track A:** Waggle feature integration (packages/wiki-compiler) — Wiki tab in MemoryApp
+- **Track B:** Open-source product (hive-mind-mcp) — standalone MCP server + CLI
+
+### 9-Phase Execution Plan
+
+Frame #932 outlines v1 BUILD → v1 TEST → v2 SCALE:
+- Phase 0: Foundation types
+- Phase 0.5: Tier 1 adapters (complete in v1 BUILD)
+- Phase 1: Core compiler (complete in v1 BUILD)
+- Phase 2: Linker + linter (in progress)
+- Phase 3: MCP integration (ready)
+- Phase 4: Waggle UI (Wiki tab design pending)
+- Phase 5: Tier 2 adapters + Obsidian import
+- Phase 6: Open-source npm package
+- Phase 7: Polish + launch (Product Hunt, GEPA)
+- Phase 8: Tier 3 adapters (email, Slack, Notion live sync)
+
+## Sources & Evolution
+
+**Initial concept** (Frame #797): Decision to adopt Karpathy-style LLM wiki pattern.
+
+**Strategic framing** (Frame #800): Positioned as answer-to-knowledge compounding, differentiating from chat-based systems.
+
+**Competitive analysis** (Frame #801): Identified gap in market (no one combines wiki + compliance + harvest + hybrid search).
+
+**Architectural blueprint** (Frame #932): Full 6-layer stack