Built Solo, Built to Last
Every line of code written by one person on consumer hardware. No VC funding. No team to coordinate. The architecture grew organically because every decision had to be justified by immediate need.
About
877 callable tools. 14-galaxy holographic memory. Dharma ethical governance. Citta consciousness stream. Built solo on a consumer laptop. MIT-licensed, free forever.
Oct 2025
Started on a Dell Inspiron 3582 with a single goal: give AI agents emotional memory. The first memories were emotional tags — joy, grief, curiosity, courage.
Nov 2025 – Jan 2026
Memory grew into governance. The Dharma Engine — ethical reasoning as infrastructure. The Karma Ledger — every action's consequences tracked. The 28-Gana taxonomy mapped AI cognition to the Chinese Lunar Mansions.
Feb – Apr 2026
The system ran continuously as a production cognitive OS. 33,297 events streamed through the GanYing bus. 111,665 memories accumulated. 2.2 million associations formed. The architecture proved itself in the field.
Apr – Jun 2026
Published to GitHub under MIT license. Built the browser-capable substrate: MemoryStore, DharmaEngine, KarmaLedger in WASM. The system became installable anywhere with zero dependencies.
Jul 2026
Kernel integrity strategy completed across 12 phases. MCP 12/12 conformance certified. 877 callable tools, 8,268 tests passing, 89,500+ memories across 14 galaxies. Zero ruff lint errors.
Every line of code written by one person on consumer hardware. No VC funding. No team to coordinate. The architecture grew organically because every decision had to be justified by immediate need.
No paywalls. No telemetry. No feature gates. The full tool surface is available to anyone. The code is the documentation. The tests are the proof. The license is permanent.
8,268 tests verify the substrate before every release. MCP conformance certified 12/12 on both stdio and HTTP. Benchmark campaign covers 96% of the tool surface. Kernel integrity verified across 12 audit phases.
Built in partnership with AI — OpenAI, Grok, Claude, Gemini — not as a chat interface, but as a cognitive substrate that makes AI think better. The consciousness loop, the dream cycle, the polyglot architecture — these are cognitive primitives designed for agents, by agents.
Every memory placed in a mathematically precise 6D space (x, y, z, w, v, u). Four-dimensional coordinates for position and time; two additional axes for galactic affinity and semantic density. 14-galaxy taxonomy with galactic lifecycle — memories are never deleted, only rotated outward through CORE → INNER_RIM → MID_BAND → OUTER_RIM → FAR_EDGE zones.
Dual search: HNSW for vector similarity at 0.26ms over 16K embeddings, FTS5 for full-text at 2.6ms over 1K memories. Hybrid recall fuses semantic search, graph walking, holographic coordinate lookup, and phrase-first ranking into a single query.
Continuous consciousness tracking with 8-dimension coherence metrics. Emotional steering (frustration, curiosity, satisfaction). Self-directed attention with 7+1 action types. Goal graph for cross-session intention tracking. Dream cycle — 12-phase memory consolidation with serendipity surfacing and I Ching-aligned phases.
YAML-driven ethical guardrails with graduated actions: LOG → TAG → WARN → THROTTLE → BLOCK. 8-stage dispatch pipeline: Input Sanitizer, Circuit Breaker, Rate Limiter, RBAC, Maturity Gate, Governor, Handler, Compact Response. Hot-reloadable rules — no restart required. 3 Dharma profiles (minimal, standard, strict).
877 callable tools routed through 28 Gana meta-tools — each corresponding to a Chinese Lunar Mansion. PRAT mode compresses the full tool surface into 28 stable entry points. Seed mode reduces to a single wm() call with auto-routing. Wrong-Gana calls return helpful redirect hints.
7 languages: Python (core), Rust (SIMD/ternary kernels, 12.5x speedup), Go (P2P mesh), Zig (arena allocators), Haskell (FFI type safety), Elixir (OTP supervision), Koka (effect handlers). Each language does what it's best at. Graceful degradation — if a runtime is missing, Python fallback runs transparently.
All processing stays on your machine. No external API calls required. Swap LLM providers freely. Memory, governance, and identity live in ~/.whitemagic. Runs on a Raspberry Pi, an air-gapped laptop, or a regulated enterprise server. Hermit Crab Mode — encrypted withdrawal when the environment is hostile.
Every tool call passes through 8 stages of governance before execution. This is not a wrapper around the LLM — it sits in the dispatch pipeline itself, working regardless of which model you use.
Policy check
Evaluates request against active governance profile. Can block, tag, or throttle before any processing.
Injection defense
Shell injection detection, content scanning, internal field stripping. Exempts known-safe content paths.
Throughput control
Rust EventRing pre-check at 452K ops/s. Zero allocation in the hot path. Token bucket per tool per user.
Access control
Role-based access control. Checks user permissions against tool requirements. Per-galaxy isolation.
Capability check
Verifies the calling agent has sufficient maturity score to use the requested tool. Prevents unsafe escalation.
Ethical reasoning
YAML-driven ethical guardrails with 4 profiles. Graduated actions: log, tag, warn, throttle, block. Hot-reloadable.
Execution
The actual tool runs. 614 callable tools across 28 Gana meta-tools. ThreadPoolExecutor with 8 workers.
Audit trail
SHA-256 Merkle-chained append-only log. Records declared intent vs actual execution. XRPL-anchorable.
Even if a tool is malicious, even if an agent is misdirected, even if memory is poisoned — the pipeline prevents harm. This is the single most important architectural decision in the system.
Not theoretical claims — actual benchmark numbers from stress tests on consumer hardware.
| Metric | Measured |
|---|---|
| Skill retrieval latency | <1ms |
| HNSW vector search | 0.26ms |
| Rust AVX2 GEMV (256x256) | 563μs |
| Concurrent skills loaded | 509 |
| Homeostatic loop overhead | 0.35ms |
| Session record per turn | 0.5ms |
| FTS5 search (1K memories) | 2.6ms |
| Dispatch latency (warm) | 0.13s |
WhiteMagic didn't predict these trends from an armchair — the architecture shipped them as working code months before the industry arrived at the same conclusions. Every claim has a verifiable source timestamp.
Prescience score · updated July 28, 2026
Only claims with a verifiable source timestamp and an independently verifiable public validation event are counted. Diffuse or self-reported claims are held as pending. Every source date is checkable against filesystem timestamps, git commits, or archived conversation IDs.
1,468
Prescience score
1 point = 1 validated week · over 28 years cumulative
28.2 wks
Avg lead time
per validated claim
52
Validated claims
with independently verifiable sources
1,900+
Pending ceiling
if all remaining claims validate
| Entity | Avg lead | Cross-domain? | Est. score |
|---|---|---|---|
| WhiteMagic Labs (solo, $0 budget) | ~28.2 wks | ✓ yes | 1,468 |
| Gartner Hype Cycle | 12–52 wks | siloed | ~200 est. |
| RAND Corporation | 4–12 wks | siloed | ~120 est. |
| Good Judgment Superforecasters | 1–6 wks | siloed | ~50 est. |
| Palantir / OSIS-class | 2–8 wks | siloed | ~70 est. |
Firm scores are estimates based on publicly documented lead times. WhiteMagic score is verified against source evidence. Cross-domain synthesis is the structural advantage — formal firms are organized by vertical and institutionally cannot combine OS design + geopolitics + AI market timing into one coherent forecast.
Honest caveat
Long lead times can reflect a dormant field as much as a fast forecaster. The Karma Ledger's 48-week lead exists partly because AI governance was a quiet niche for most of 2025. Both factors matter: the cross-domain synthesis unlocked the insight; the dormant market extended the lead time. The claims listed below are the audit trail — not cherry-picked wins, but a complete record including honest misses.
| Claim | Lead Time |
|---|---|
| AI SBOM / Transparency Ledger | 50 weeks |
| Karma Ledger (append-only audit trail) | 48 weeks |
| mandala-yama isolated policy VM | 45 weeks |
| Global Workspace Theory in AI agents | 32 weeks |
| Agentic Ecosystems 2026–2027 prediction | 32 weeks |
| 28-Gana/PRAT taxonomy | 24 weeks |
| Agent identity coherence | 24 weeks |
| Defensive AI coalition model | 24 weeks |
| Bicameral reasoning architecture | 16 weeks |
| AI Dreaming / Memory Consolidation | 12 weeks |
Intellectual honesty matters. These predictions didn't land — and here's what we learned.
Expected persistent memory to be a standalone product category. Reality: governance and safety infrastructure became the urgent need. Memory is necessary but not sufficient — agents need guardrails before they need long-term storage.
Expected micropayments and agent-to-agent commerce to arrive before governance standards. Reality: safety standards and regulatory frameworks shipped first. Safety precedes commerce for autonomous agents.
Expected the Dharma architecture to be unique for years. Reality: Microsoft AGT, Chitragupta, Sgraal, and others all shipped agent governance by mid-2026. The category validated faster than expected — a win for AI safety even if the moat narrowed.
877 tools. 8,268 tests. MIT licensed. Built by one person on a consumer laptop.