FORGE

3D architecture builder + multi-AI workshop — the workshop where AI works.

---

Concept

Open any project — a codebase, a network topology, a paper, a Notion workspace — and FORGE renders it as a navigable 3D world. Inside that world, AI agents (Claude, GPT, Gemini) read the modules, analyze them, fix them, test them, and certify them. The deliverable is ADN signé (signed digital DNA): a cryptographically signed audit artifact that captures the project's architecture, findings, and resolution chain.

FORGE is not a renderer. It is the workshop — the cage where AI does work that humans then review.

What ships today (v1.0.3-beta)

• Phase J — Productive Modules: modules don't just visualize, they live and work in 3D. Each module exposes a 2D context pane for productive interaction.
• Phase K — Universal scanner (L2-L7): ARP + NDP + mDNS at L2, ICMPv4/v6 + traceroute at L3, opt-in TCP/UDP at L4, HTTP + TLS + SSH + SNMP at L7, plus topology builder, benchmarks, and end-to-end integration tests.
• Phase H — Security Workbench: vulnerability surfaces rendered as glowing 3D modules, NVD CVE pipeline, threat-model overlays.
• Phase N + Foreman + Q + R + I + L + M + SEED integration: the orchestration spine — intent → plan → dispatch → verify, with persistence, branding, hierarchical navigation, and the SEED DSL feeding compact briefs into agent context windows.
• Multi-AI live in 3D: Claude API wired into the Messenger panel (P0.4); audit pipeline auto-generates findings as glowing 3D modules with notes, marks, and resolvedBy links.
• Self-audit milestone: the first audit of FORGE was generated by Claude, running inside FORGE v1.0.3-beta live, against its own seven-pillar audit framework.

Stats

• 2,871 / 2,871 tests green at end of marathon session (up from 1,956 mid-cycle).
• 80+ files Phase 2-5 · 109 files Phase J · ~110 files / ~5,800 lines Phase K.
• 22+ parallel AI agents per MEGA session, executor pattern, isolated worktrees.
• DPAPI SecretsVault + AES-CBC → AES-GCM crypto migration completed.

Long-term vision — "Kubernetes for AI agents"

Five principles guide the post-Mythos roadmap (Phases P → Q → R → S):

1. Constitution lives in FORGE, not in the model. Behavioral guarantees are platform-level.
2. IAgentRuntime abstraction. Swap Claude / GPT / Gemini / on-prem models without rewriting workflows.
3. Open standards over proprietary lock-in. .dna artifacts, signed audit chains, open exchange formats.
4. RAG over fine-tune. Fresh project context beats stale weight surgery.
5. Eval harness as first-class citizen. Every agent action is measured against a regression suite.

Stack

Godot 4 (rendering, scene graph, input) · C# class libraries (Forge.Core, Forge.Godot, Seed.Engine DLLs) · Python integration layer · WebSocket transport · NetGuard suite back-end integration · Claude / OpenAI / Gemini APIs · pytest + xUnit · GitHub Actions CI.

Status & licensing

• v1.0.3-beta, single-developer build, audited live by its own embedded Claude.
• Source code private. This repository hosts architecture, screenshots, and design notes only.
• Pricing target (enterprise): $75K – $150K per year. Signed-DNA audit deliverable: $50K+.
• Target market: Thales-class enterprise — Canadian cybersecurity, aerospace, defense, regulated AI.
• Open to demos, partnership conversations, and pilot programs.

Contact

sxc3030@gmail.com · github.com/sxc3030-eng

See ARCHITECTURE.md for the project ingestion and phase-stack diagrams.

---

_{© 2026 Simon Cantin}