moving

docs/wiki-test/concepts/development-velocity.md

@@ -0,0 +1,92 @@
+---
+type: concept
+name: Development Velocity
+confidence: 0.95
+sources: 14
+last_compiled: 2026-04-13
+---
+
+# Development Velocity
+
+## Summary
+
+This page synthesizes patterns across 20 development sessions to understand
+how [[Waggle OS]] actually gets built — what works, what doesn't, and what
+the data says about productivity.
+
+No single session handoff contains this analysis. It emerges from looking
+across all of them.
+
+## Velocity by Phase
+
+| Period | Sessions | Commits | Key Output |
+|--------|----------|---------|------------|
+| Mar 27-31 | 5 | ~10 | M2 sprint: P0 fixes, resilience, cleanup |
+| Apr 1-5 | 4 | ~8 | Deep refactoring, multiple audit rounds |
+| Apr 6-7 | 2 | ~6 | Agent intelligence sprint (9 features, 112 tests) |
+| Apr 8 | 1 | 6 | Structural refactors (chat.ts, personas.ts split) |
+| Apr 9-10 | 2 | 11 | ULTRAPLAN Phases 1-9 (120 files, 17K lines) |
+| Apr 11 | 1 | 17 | Phases A+B (12 feature + 5 polish commits) |
+| Apr 12 | 5 | ~30 | Phases C+D, E2E, tier restructure, binary |
+| Apr 13 | 2 | 33 | Entire backlog + Memory MCP + Wiki Compiler spec |
+
+**Total: ~20 sessions, ~120 commits, 18 days from "all broken" to backlog cleared.**
+
+## What Accelerated
+
+1. **CLAUDE.md as contract.** Once the authoritative CLAUDE.md was in place
+   (around Apr 8), sessions became dramatically more productive. The agent
+   could self-direct against a clear spec instead of guessing.
+
+2. **"You lead" delegation.** Marko's shift to *"yes whatever you choose you
+   lead"* (Apr 12-13) enabled marathon autonomous sessions. 29 commits in a
+   single session wouldn't happen with approval on every change.
+
+3. **Memory handoffs.** Session handoff memory files let each new session pick
+   up exactly where the last left off. No ramp-up time. The first message is
+   "continue" and work starts immediately.
+
+4. **Playwright over manual testing.** After Marko burned out on manual binary
+   verification (Apr 12), switching to headless Playwright E2E removed the
+   human bottleneck. 96 E2E tests replaced "click here, does it work?"
+
+## What Slowed Down
+
+1. **The "all broken" period (Apr 1).** Multiple restart cycles, CORS errors,
+   vault invisible, agents returning "local mode." Lost ~half a day to
+   debugging basic wiring issues.
+
+2. **Audit fatigue (Apr 1).** Marko asked for 4+ full codebase audit rounds
+   in a single day. Each audit found real issues, but diminishing returns
+   set in. The fix: ship and iterate, not audit to perfection.
+
+3. **Context resets.** New sessions without clear handoffs required re-reading
+   the entire codebase state. The memory system solved this but took until
+   Apr 8 to be fully operational.
+
+4. **Binary build + test cycle.** Building the Tauri binary takes minutes.
+   Testing it manually takes more. This blocked the feedback loop until
+   Playwright replaced manual testing.
+
+## The Mega Session Pattern
+
+The most productive sessions follow a pattern:
+1. Clear handoff note from previous session
+2. Marko says "continue" or gives broad direction
+3. Agent reads state, makes a plan, executes autonomously
+4. Marko checks in periodically for strategic steering
+5. Session ends with detailed handoff note
+
+The April 13 mega session (29 commits) is the purest example. Marko gave
+broad directives ("do team server", "compose new docker", "go for ollama")
+and the agent executed full feature implementations autonomously.
+
+## Contradiction
+
+The April 1 "audit everything" approach vs. the April 13 "you lead" approach
+represent opposite strategies. Both produced results, but the "you lead"
+approach produced 3-5x more output per session. The audit approach caught
+real bugs but also burned time on diminishing-return passes.
+
+**Resolution:** The optimal pattern is likely: "you lead" for implementation,
+followed by one focused review pass (not 4).

docs/wiki-test/concepts/mind-architecture.md

@@ -0,0 +1,59 @@
+---
+type: concept
+name: Mind Architecture
+confidence: 0.96
+sources: 6
+last_compiled: 2026-04-13
+related_entities:
+  - Waggle OS
+---
+
+# Mind Architecture
+
+## Summary
+
+The .mind file is [[Waggle OS]]'s core technical innovation — a single SQLite
+database per user or workspace that stores persistent memory in a 6-layer
+architecture inspired by video codecs (I-Frames and P-Frames).
+
+## The 6 Layers
+
+| Layer | Name | Content | Load Time | Size |
+|-------|------|---------|-----------|------|
+| 0 | Identity | WHO the agent is (name, role, personality) | <1ms | ~500 tokens |
+| 1 | Awareness | WHERE the agent is (active tasks, context) | <50ms | ~2000 tokens |
+| 2 | Memory Stream | I-Frames (snapshots) + P-Frames (deltas) | <200ms | Unbounded |
+| 3 | Knowledge Graph | Entities + typed relations + confidence | Query-time | Unbounded |
+| 4 | Session Store | Frame grouping by session/conversation | Query-time | Unbounded |
+| 5 | Embedding Index | sqlite-vec vectors for semantic search | Query-time | ~50MB |
+
+## How Search Works
+
+Hybrid search combining three strategies with Reciprocal Rank Fusion (RRF):
+1. **BM25** via FTS5 — catches exact keyword matches
+2. **Vector similarity** via sqlite-vec — catches semantic meaning
+3. **Knowledge Graph traversal** — catches relationship-based connections
+
+## Key Design Decisions
+
+- **SQLite, not Postgres:** Zero external dependencies. Embedded. Portable.
+  A .mind file IS the memory — copy it, and the agent's memory travels with it.
+- **Video codec metaphor:** I-Frames are complete snapshots (like keyframes).
+  P-Frames are deltas/updates. This enables efficient storage with temporal
+  reconstruction.
+- **Confidence scoring:** Every entity and relation has a confidence score
+  (0-1). Multiple sources reinforcing the same fact increase confidence.
+  Contradictions lower it.
+- **Frame sources:** Every frame is tagged: user_stated, tool_verified,
+  agent_inferred, system, import. This enables trust hierarchies.
+
+## Scale
+
+Target: ~500MB-1GB per .mind file per employee. Current personal.mind has 202
+frames and 2,734 entities. Production workloads expected to reach thousands
+of frames with sub-200ms query times.
+
+## Relations
+
+- [[Waggle OS]] — the product built on this architecture
+- [[LLM Provider Stack]] — how the agent routes LLM requests for memory operations

docs/wiki-test/entities/egzakta-group.md

@@ -0,0 +1,58 @@
+---
+type: entity
+entity_type: organization
+name: Egzakta Group
+confidence: 0.93
+sources: 5
+last_compiled: 2026-04-13
+related_entities:
+  - Waggle OS
+  - KVARK
+  - Marko Markovic
+---
+
+# Egzakta Group
+
+## Summary
+
+Egzakta Group is a EUR 31M technology group founded in 2005 by three partners:
+[[Marko Markovic]], Zoran Radisavljevic, and Nenad Tesic. The holding company
+(Egzakta Holding BV) is domiciled in the Netherlands for EU positioning.
+The group has 200+ professionals and targets 400+ by 2028.
+
+## Key Facts
+
+- **Founded:** 2005 (19 years of operation)
+- **Founders:** Marko Markovic, Zoran Radisavljevic, Nenad Tesic
+- **Holding:** Egzakta Holding BV (Netherlands)
+- **Team size:** 200+ professionals, targeting 400+ by 2028
+- **Revenue:** EUR 31M normalized (2025)
+- **Presence:** European expansion in progress
+
+## Strategic Positioning
+
+The group operates a 4-layer technology stack. LinkedIn content and product
+promotion must reflect the REAL scale and ambition — not a solo consultant,
+but a founding partner of a EUR 31M technology group with European expansion
+ambitions.
+
+## Products
+
+- [[KVARK]] — Sovereign enterprise AI platform (EUR 1.2M contracted)
+- [[Waggle OS]] — AI agent platform (demand-gen for KVARK)
+
+## Business Model (for Waggle)
+
+Decided 2026-03-06 during brainstorm:
+- **Open core** (like GitLab, Supabase, Cal.com)
+- **Open source:** agent core, MCP framework, skill engine, basic clients
+- **Proprietary:** GEPA engine, Memory Weaver, multi-tenant server, admin, SSO
+- NOT just for companies — individuals, students, one-man shops too
+- "Like Office 365: both personal and enterprise"
+- Commercial, fee-based, cheap, go for volumes
+
+## Relations
+
+- [[Waggle OS]] — AI platform product, demand-gen engine
+- [[KVARK]] — Enterprise AI platform, EUR 1.2M contracted
+- [[Marko Markovic]] — Co-founder and primary builder

docs/wiki-test/entities/kvark.md

@@ -0,0 +1,60 @@
+---
+type: entity
+entity_type: project
+name: KVARK
+confidence: 0.91
+sources: 8
+last_compiled: 2026-04-13
+related_entities:
+  - Egzakta Group
+  - Waggle OS
+  - EU AI Act Compliance
+---
+
+# KVARK
+
+## Summary
+
+KVARK is [[Egzakta Group]]'s sovereign enterprise AI platform — the top of the
+Waggle funnel. It deploys AI infrastructure on customer premises with full data
+sovereignty, governance, and audit trails.
+
+## Key Facts
+
+- **Revenue:** EUR 1.2M contracted
+- **Product site:** https://www.kvark.ai
+- **Positioning:** "Everything Waggle does — on your infrastructure, connected
+  to all your internal systems. Full data pipeline injection, your permissions,
+  complete audit trail, governance. Your data never leaves your perimeter."
+- **Target:** Enterprise customers who need sovereign AI deployment
+- **Compliance angle:** DORA + NIS2 + AI Act + GDPR convergence
+
+## Funnel from Waggle
+
+```
+Waggle Free → Pro ($19) → Teams ($49/seat) → KVARK (consultative sale)
+```
+
+[[Waggle OS]] is the demand-creation and qualification engine. Users experience
+AI-native work individually, upgrade to teams for institutional knowledge,
+then need enterprise governance → KVARK.
+
+## Enterprise Pitch
+
+> "Every AI interaction your organization has — across email, Slack, meetings,
+> code reviews, customer calls — automatically captured in a living knowledge
+> base. Your people get a second brain. Your compliance team gets an AI Act
+> audit trail. With KVARK: all on YOUR infrastructure."
+
+## Technical Integration
+
+- `kvark-tools.ts` in Waggle agent package: `kvark_search`, `kvark_ask_document`
+- Gated to Business/Enterprise tiers
+- License server: https://license.waggle-os.ai/validate
+- SaaS cloud: https://cloud.waggle-os.ai
+
+## Relations
+
+- [[Egzakta Group]] — parent company
+- [[Waggle OS]] — demand-gen funnel
+- [[EU AI Act Compliance]] — compliance-by-default strengthens KVARK pitch

docs/wiki-test/entities/marko-markovic.md

@@ -0,0 +1,81 @@
+---
+type: entity
+entity_type: person
+name: Marko Markovic
+confidence: 0.97
+sources: 20
+last_compiled: 2026-04-13
+---
+
+# Marko Markovic
+
+## Who He Is
+
+Co-founder of [[Egzakta Group]] (EUR 31M, 200+ people, founded 2005) alongside
+Zoran Radisavljevic and Nenad Tesic. Primary developer and product owner of
+[[Waggle OS]]. Strategic lead on [[KVARK]].
+
+Based in Serbia. EU presence via Netherlands holding (Egzakta Holding BV).
+
+## What He's Building
+
+| Project | Role | Status |
+|---------|------|--------|
+| [[Waggle OS]] | Sole developer + product owner | Near launch, backlog cleared |
+| [[KVARK]] | Strategic lead | EUR 1.2M contracted |
+| The AI Sovereignty Report | Author | LinkedIn newsletter, launched Apr 2026 |
+| Cortex | Architect | Concept — personal intelligence cockpit |
+| ReFarm Discovery | Project lead | Delivered — 9.6 MB proposal for vertical farming |
+| KRA Proposal | Proposal author | Kenya Revenue Authority, World Bank procurement |
+
+## LinkedIn Presence
+
+- **Followers:** 4,521
+- **Impressions:** 340K+
+- **Viral posts:** AI Compliance Officer (26 reshares, 11 comments), 5,600 Startups (246 reshares, 47 comments)
+- **Auto-posting:** Maton.ai gateway + GitHub Actions cron (repo: marolinik/SocialPresence)
+- **Newsletter:** "The AI Sovereignty Report" — LinkedIn's algorithm heavily boosts newsletters. All 4,521 followers auto-invited on creation.
+
+## How He Works (observed across 422 messages, 20 sessions)
+
+**Decision-making:** Fast, opinionated, delegates execution. Typical pattern:
+states a direction in 1-2 sentences, expects the agent to plan and execute.
+Examples: *"yes whatever you choose you lead"*, *"do team server, you have
+all docker running"*, *"compose new docker, kill the old one."*
+
+**Marathon sessions:** Prefers long, intensive sessions over many short ones.
+The April 12-13 stretch was 29 commits in a single session. Multiple sessions
+run 6-12 hours.
+
+**Quality bar:** Demands deep audits. *"one more turn pls again using
+CLAUDE.md on rules, whole code"* — asked for 4+ full codebase audit rounds
+in a single day (April 1). Wants "perfect UI/UX" — not just functional.
+
+**Frustration triggers:** Broken basic functionality (*"fuck all broken"* when
+the agent returned "local mode" errors). Long waits with no progress
+(*"what is happening 44 mins..."*). Having to click things manually when the
+agent should be leading via Playwright.
+
+**Strategic thinking:** Regularly jumps from implementation to big-picture.
+Mid-session will pivot to topics like EU AI Act compliance, Karpathy's LLM
+Wiki, newsletter strategy, KVARK enterprise pitch. Expects the agent to
+context-switch fluidly.
+
+**Hardware:** Windows laptop with 2 GPUs (NVIDIA RTX 5070, 8GB), Intel Ultra 9
+285H, 64GB RAM. Asked about GPU memory ranges and which laptop has the largest
+GPU (March 28).
+
+## Key Decisions Made
+
+| Date | Decision | Context |
+|------|----------|---------|
+| Mar 6 | Name: "Waggle" | Brainstorm with bee metaphor mapping |
+| Mar 6 | Open core model | Like GitLab/Supabase — open agent core, proprietary GEPA/Weaver |
+| Mar 6 | Not just enterprise — individuals, students too | "Like Office 365: both personal and enterprise" |
+| Apr 1 | Full refactor after "all broken" | Multiple audit rounds, strict CLAUDE.md adherence |
+| Apr 10 | Memory Harvest is P0 | "Bring Your Memory Home" as primary enterprise hook |
+| Apr 12 | Tier restructure | Trial 15d → Free → Pro $19 → Teams $49 → KVARK |
+| Apr 12 | Memory + Harvest free forever | Moat strategy — never paywall the core value |
+| Apr 13 | Wiki Compiler spec | Karpathy extension, open-source dual track |
+| Apr 13 | Universal source pipeline | Full second brain, not just AI agent memory |
+| Apr 13 | EU AI Act as compliance layer | Second brain IS the audit trail |

docs/wiki-test/entities/waggle-os.md

@@ -0,0 +1,100 @@
+---
+type: entity
+entity_type: project
+name: Waggle OS
+confidence: 0.98
+sources: 25
+last_compiled: 2026-04-13
+---
+
+# Waggle OS
+
+## What It Is
+
+A desktop AI agent platform where agents have persistent memory, communicate
+via the Waggle Dance protocol, and work across isolated workspaces. Ships as
+a Tauri 2.0 binary (Windows + macOS) with a React frontend and Node.js sidecar.
+
+Strategic function: demand-creation engine for [[KVARK]]. Users experience
+AI-native work individually, get hooked on persistent memory, upgrade to teams,
+then need sovereign deployment.
+
+Named after the waggle dance — how bees communicate location of food sources.
+Chosen March 6, 2026. Runners-up: Hivemind, Apis, Mycel.
+
+## Tech Stack
+
+- **Frontend:** React 18 + TypeScript + Vite + Tailwind + shadcn/ui
+- **Desktop shell:** Tauri 2.0 (Rust)
+- **Backend:** Fastify sidecar (Node.js, bundled into Tauri)
+- **Memory:** SQLite via better-sqlite3 + sqlite-vec (see [[Mind Architecture]])
+- **Auth:** Clerk (JWT)
+- **Design:** Hive DS — honey #e5a000, hive-950 #08090c, accent #a78bfa
+- **LLM routing:** 3-layer stack (see [[LLM Provider Stack]])
+
+## What's Built (verified April 13)
+
+| Feature | Status |
+|---------|--------|
+| 22 agent personas | Shipped |
+| 80+ agent tools | Shipped |
+| 148 MCP connectors (visual catalog) | Shipped |
+| 6-layer .mind memory architecture | Shipped |
+| Memory Harvest (ChatGPT, Claude, Gemini, Claude Code) | Shipped |
+| Knowledge Graph (typed entities + relations) | Shipped |
+| HybridSearch (FTS5 + vector + RRF) | Shipped |
+| Per-workspace isolation + cross-workspace search | Shipped |
+| Shared team memory (push-on-write, pull-on-activate) | Shipped |
+| S3/MinIO storage for team files | Shipped |
+| Tiered autonomy (Normal/Trusted/YOLO) | Shipped |
+| The Room (multi-window workspace with per-window personas) | Shipped |
+| Trial expiry modal + onboarding auto-trial | Shipped |
+| Budget hard cap (soft warn + hard stop) | Shipped |
+| Ollama/local inference (hardware scan + model recs) | Shipped |
+| Docker stack (Postgres, Redis, MinIO, LiteLLM) | Shipped |
+| 40/40 E2E tests passing | Shipped |
+| Memory MCP plugin (14 tools, 5 fixes applied) | Shipped |
+
+## Monetization
+
+See [[Tier Strategy]] for full details.
+
+Trial 15d (everything) → Free (5ws, agents) → Pro $19 → Teams $49/seat → KVARK.
+Memory + Harvest free forever as moat.
+
+## The Journey (what actually happened)
+
+**The ugly start (Mar 27 - Apr 1):** Backend broken, CORS issues, vault
+invisible, agents returning "local mode" errors. Marko: *"fuck all broken."*
+Multiple restart-the-server cycles. Deep audit rounds to find unwired code.
+
+**The grind (Apr 1-7):** Multiple full-codebase audit passes. UI/UX overhaul
+with 12-persona simulation and 48-gap analysis. Agent intelligence sprint —
+9 features, 112 tests. Persona expansion to 22.
+
+**The acceleration (Apr 8-10):** Structural refactors (chat.ts, personas.ts
+split). ULTRAPLAN Phases 1-9 in a single mega session (11 commits, 120 files,
+17K lines). First real Memory Harvest — 156 frames from Claude Code sessions.
+
+**The killer sprint (Apr 11-12):** Phases A+B (The Room, filesystem, tiered
+autonomy) in one day. Binary verification — 11 bugs found and fixed. E2E
+suite green. Phases C+D. Tier restructure. Binary built.
+
+**The mega finish (Apr 13):** 29 commits. Entire P0-P4 backlog cleared. Team
+memory, S3 storage, global KG, budget cap, Ollama, Memory MCP plugin — all
+shipped in one session. Dead `app/src/` removed (13K LOC, 77 files).
+
+## What's Left
+
+1. **Stripe products** — Marko creates PRO + TEAMS in Stripe dashboard
+2. **Code signing** — purchase certificate for Windows installer
+3. **Wiki Compiler** — spec complete (1,639 lines), not yet built
+4. **GEPA prompt optimization** — enable smaller models matching Opus quality
+
+## Known Architectural Decisions
+
+- **Open core model** (like GitLab, Supabase) — open: agent core, MCP, skills.
+  Proprietary: GEPA, Memory Weaver, multi-tenant server, admin, SSO.
+- **Desktop-first, not web-first** — Tauri for native performance + local data
+- **SQLite not Postgres for personal data** — zero deps, portable, .mind file IS the memory
+- **Memory + Harvest free forever** — moat strategy, never paywall the core value

docs/wiki-test/health.md

@@ -0,0 +1,91 @@
+---
+type: health
+last_compiled: 2026-04-13
+---
+
+# Wiki Health Report
+
+## Source Quality Assessment
+
+| Source | Files | Quality | Notes |
+|--------|-------|---------|-------|
+| Memory handoffs | 12 | Excellent | Curated, structured, timestamped |
+| Strategy memories | 5 | Excellent | Decisions with rationale |
+| Feature memories | 4 | Good | Technical specs, architecture |
+| Feedback memories | 1 | Good | Clear behavioral guidance |
+| Session JSONL files | 20 | Raw | 422 user messages, needs extraction |
+| personal.mind frames | 202 | Poor | 80% E2E test pollution |
+| personal.mind KG | 2,734 entities | Very Poor | 80% misclassified noise |
+
+**Verdict:** Memory files are the gold standard. Session JSONLs have real
+decisions but need filtering. personal.mind is polluted from automated testing
+and should NOT be used as a compilation source until cleaned.
+
+## Gaps — Topics That Should Have Pages But Don't
+
+### Critical (referenced in 5+ sources, no synthesis)
+- **Memory Harvest Pipeline** — core feature, mentioned everywhere, no dedicated page
+  explaining what adapters exist, what the UX flow is, what's imported vs. not
+- **Docker Stack** — Postgres, Redis, MinIO, LiteLLM all in docker-compose.yml
+  but no page explaining what each does and how they wire together
+- **Agent Architecture** — 22 personas, 80+ tools, but no synthesis page on
+  how the agent loop works, how personas are selected, how tools are filtered
+
+### Important (referenced in 2-4 sources)
+- **Onboarding Flow** — 7-step wizard, revised multiple times, no synthesis
+- **Design System (Hive DS)** — tokens, colors, components scattered across sessions
+- **E2E Testing Strategy** — evolved from manual to Playwright, no summary
+
+### Missing Relations
+- Marko ↔ Zoran Radisavljevic (co-founders, never linked)
+- Marko ↔ Nenad Tesic (co-founders, never linked)
+- Waggle OS ↔ Tauri 2.0 (core dependency, not an entity)
+- KVARK ↔ DORA/NIS2 (compliance regulations that strengthen the pitch)
+
+## Contradictions
+
+### Entity Count Evolution
+- Mar 17: "13 personas, 28 connectors"
+- Apr 9: "22 personas, 102 MCP servers"
+- Apr 10: "22 personas, 148 MCP servers"
+- Apr 13: "22 personas, 148 MCP connectors"
+**Status:** Not a contradiction — shows growth over time. Timeline captures this.
+
+### Test Count
+- Mar 23: 298/299 tests
+- Apr 7: 4,409 tests (after agent intelligence sprint)
+- Apr 12: 96 E2E tests (separate suite)
+- Apr 13: 40/40 E2E (trimmed to stable subset)
+**Status:** Different test suites measured at different times. Clarify in
+[[Waggle OS]] page which suites are which.
+
+### Tier Pricing
+- Original: Solo/Teams/Business/Enterprise (4 tiers)
+- Apr 12 restructure: Trial/Free/Pro $19/Teams $49/KVARK (5 tiers)
+**Status:** Resolved — new structure is canonical.
+
+## Data Quality Recommendations
+
+1. **Clean personal.mind KG** — delete entities with names matching common
+   nouns/verbs ("Begin Week", "High Risk", etc.). Dedup by normalized name.
+   This is prerequisite for any automated wiki compilation.
+
+2. **Extract decisions from session JSONLs** — Marko's messages contain
+   real strategic decisions ("memory + harvest free forever", "open core
+   model", "not just enterprise"). These should be first-class decision
+   frames, not buried in chat logs.
+
+3. **Link memory files to git commits** — each handoff references commit
+   hashes. Cross-linking would enable "what was built in this session" →
+   actual code changes.
+
+## Compilation Statistics
+
+| Metric | Value |
+|--------|-------|
+| Memory files used | 29 |
+| Session JSONLs scanned | 20 (422 user messages) |
+| Pages compiled | 7 |
+| Pages that should exist | ~25-30 |
+| Compilation time | ~3 minutes (manual, this test) |
+| Estimated automated cost | ~$0.08 (Haiku, 7 pages) |

docs/wiki-test/index.md

@@ -0,0 +1,51 @@
+# Marko's Second Brain — Wiki Index
+
+*Compiled from 29 memory files + 422 session messages + git history.*
+*Source: Claude Code sessions, March 27 — April 13, 2026.*
+*Last compiled: 2026-04-13*
+
+---
+
+## Organizations
+- [[Egzakta Group]] — EUR 31M tech group, 200+ people, 3 founders, Netherlands holding
+
+## Products
+- [[Waggle OS]] — Desktop AI agent platform with persistent memory
+- [[KVARK]] — Sovereign enterprise AI platform (EUR 1.2M contracted)
+
+## People
+- [[Marko Markovic]] — Co-founder, Egzakta. Builder of everything here.
+
+## Concepts
+- [[Mind Architecture]] — 6-layer video-codec memory in SQLite (.mind files)
+- [[Tier Strategy]] — Trial 15d → Free → Pro $19 → Teams $49 → KVARK
+- [[EU AI Act Strategy]] — Compliance by default, Aug 2 2026 deadline
+- [[Memory Harvest]] — Import AI conversations from any platform
+- [[Wiki Compiler]] — Karpathy LLM Wiki + Waggle memory = second brain (THIS)
+
+## Development Timeline
+
+| Date | What Happened | Sessions |
+|------|--------------|----------|
+| Mar 6 | Name chosen: "Waggle" (from waggle dance). Vision, business model. | Brainstorm |
+| Mar 11 | Phase 3 gap analysis — 48 critical gaps identified | Analysis |
+| Mar 17 | B2B lead gen stress test — agent speed/persistence gaps found | Testing |
+| Mar 23 | UAT score 84/100, "Honey on Dark Steel" design system | M2 |
+| Mar 27-31 | M2 sprint — 13 sessions, 10 commits. P0 fixes, network resilience, TS cleanup, keyboard flow, telemetry | M2 Sprint |
+| Apr 1-5 | Deep refactoring marathon — multiple audit rounds, UI/UX overhaul | Refactor |
+| Apr 6-7 | Agent intelligence sprint — 9 features, 112 tests, persona expansion | Phase 1 |
+| Apr 8 | Structural refactors — chat.ts split, personas.ts split, tools extract | Session 2 |
+| Apr 9-10 | ULTRAPLAN Phases 1-9 complete. 148 MCP servers. Real harvest (156 frames). | Mega session |
+| Apr 11 | Phases A+B: The Room, per-session orchestrators, tiered autonomy. 50 tests. | Killer day |
+| Apr 12 (AM) | Phase A/B binary verification. 11 bugs fixed. | Verification |
+| Apr 12 (PM) | E2E suite 13/13 green. System map. Loading shimmer. | Polish |
+| Apr 12 (EVE) | Phases C+D. 96 E2E. Tier restructure. Binary built. | Mega session |
+| Apr 12 (NIGHT) | 3 P0s fixed. WeaverPanel. 101 dirty files committed. | Cleanup |
+| Apr 12 (S2) | Trial expiry modal. MultiMindCache. Cross-workspace search. | Features |
+| Apr 13 | **29 commits.** Team memory, S3/MinIO, KG, budget cap, Ollama, Memory MCP. Entire backlog cleared. | MEGA |
+| Apr 13 (now) | Memory MCP 5 fixes. Wiki Compiler spec (1,639 lines). | This session |
+
+## Open Items
+- **Stripe:** Create PRO ($19) + TEAMS ($49) products in dashboard
+- **Code signing:** Purchase certificate for Windows installer
+- **Wiki Compiler:** 1,639-line spec ready, not yet implemented