DeckerGUI Progress Update: From 3-DGM Governance to a Token-Governed Agentic Ecosystem

Wan Mohd Azizi bin Wan Hosen / Rikayu Wilzam — Sarawak, Malaysia
deckergui.my | app.deckergui.my
Progress Update · August 2026

📄 Full PDF: DeckerGUI Progress Update

Abstract

We report on the current progress of DeckerGUI, a unified agentic ecosystem governed by a 3-DGM (Three-Dimensional Governance Model) gate and oriented around token-governed agent maintenance. Since the DGUI Emitter protocol established the separation of governance from execution, we have shipped a live emitter console, a companion Enterprise KPI Tokenizer for progression tracking, a public landing surface, and production tunneling infrastructure. Token Trace reproduces a measured 88.11% saving on a representative docking flow (5,325 tokens/$3.20 baseline versus 633 tokens/$0.38 emitter-optimized), while the structural analysis holds at a 99.4% reduction for docking agents once execution is shifted to metered Claw keys.

We are now building the DGM (Digital Guild Master) maintenance layer: per-feature Technician agents on the opencode harness, a Monday maintenance scheduler, first-time-use prerequisite setup, and a cross-platform DGM MCP server. We clarify the relationship between DGM and MCP: they overlap only on the callable-tool surface, while DGM's substance — the gate, digital keys, central seeder ledger, token metering, and maintenance lifecycle — has no MCP equivalent.

Keywords: DeckerGUI, DGM, DGUI Emitter, Agentic Docking, Token Governance, KPI Tokenizer, MCP, Agent Maintenance

1. Introduction

The DeckerGUI ecosystem introduces a 3-DGM (Three-Dimensional Governance Model) Agentic Docking Gate that enforces security, capability, and policy compliance for every agent operation. Each operation passes through three sequential layers — DSYNC Memory Reset (threat scanning), Capability Guard (authorization), and Policy Supervisor (compliance) — before seeding is permitted, and every outcome is recorded exclusively through the Central YoloMoE Seeder, the single legitimate write path in the system.

A critical observation emerged during implementation: docking agents, after passing the gate, continued to consume tokens during their execution phase. Governance was coupled to execution inside the same agent lifecycle. The DGUI Emitter protocol eliminated this coupling: after gate clearance, a docking agent emits only a lightweight ClawIntent and steps aside, while a specialised Claw agent performs the actual work under a blockchain-style Digital API Key with scoped permissions, a remaining quota, and an expiry timestamp.

This document is a progress update rather than a formal specification.

2. What We've Shipped

2.1 DGUI Emitter Console

Deployed live at emitter.deckergui.my (dev: localhost:3007, API: localhost:3006). Sixteen docking features across eight main-feature views (DSYNC, DGUI Persona, DGM Factory, DGUI Market, DGUI Tufty, DGUI Hub Network, DGUI DOCS, DGUI AGENT) plus a SYSTEM group (Connection, Seed Status). Every sub-feature runs the strict 3-DGM gate against the live backend — currently 16 registered features and a live seed history of 50 entries.

Service	Port	Public URL	Status
DGUI Emitter API	3006	emitter.deckergui.my	up (16 features, 50 seeds)
DGUI Emitter console	3007	emitter.deckergui.my	up
KPI Tokenizer API	3004	app.deckergui.my	up (v1.1.0)
KPI Tokenizer UI	3003	app.deckergui.my	up
DGUI landing page	3010	landing.deckergui.my	up (HTTP 200)
Cloudflare tunnel	—	tunnel deckergui-api	up (3 replicas)

2.2 Token Trace and the Measured 88.11%

Token Trace reproduces the token economics of a representative docking flow in the browser: against a baseline of 5,325 tokens / $3.20, the emitter-optimized flow consumes 633 tokens / $0.38 — a measured reduction of 88.11%. The remaining consumption is deliberately retained governance (gate, intent construction, result logging) — the cost of staying auditable.

2.3 3-DGM Gate, DSYNC, Digital API Keys, and Settlement

• DSYNC Memory Reset scans agent memory for nine threat patterns (eval(), new Function, __proto__, constructor[, process.env, child_process, …), quarantining matches with severity classification.
• Digital API Keys carry keyHash, ownerAgentId, scopes, remainingQuota, expiresAt, rewardEligible; minted per feature, filled into agentic slots, revoked on disconnect.
• ClawIntent contracts are signed, nonced, and carry maxDockingTokens: 0 — execution budget is impossible at the type level.
• Settlement and metering record every execution with the claw key hash and tokens used.

2.4 Enterprise KPI Tokenizer (Progression Tracking)

Runs at app.deckergui.my (dev: localhost:3003, API: localhost:3004): Dashboard, Employee views, Model assignment, Quota configuration, KPI Report, Log Usage, and API Key management. Model type/source is auto-detected from the model string. The budget engine maps salary to monthly tokens at 1,000 tokens per RM across four tiers (Entry, Mid, Senior, Executive) and a 14-model pricing catalog. Thirteen API endpoints including usage query, audit trail, salary-to-token conversion, and the AI-provider proxy.

The KPI Tokenizer is the progression store: after a session or a finished project, the DGM layer will prompt you to review your progression — by day, by model, by harness, by tool — on the Enterprise KPI Tokenizer.

2.5 Landing Surface and Tunneling Infrastructure

landing.deckergui.my (dev: localhost:3010) over HTTPS via the Cloudflare named tunnel deckergui-api (three replicas, metrics ports 4100/4101/4102). All public surfaces verified reachable.

3. The Numbers

Phase	Standard	Emitter	Note
3-DGM Gate (governance)	50	50	unchanged
Intent construction	10	< 1	reduced > 90%
Execution	10,000	10,000	shifted to Claw key
Result logging	5	5	unchanged
Docking agent total	10,065	56	−99.4%
Claw agent total	0	10,000	separate metered budget

The measured browser reproduction narrows this to 88.11% on a realistic mixed flow (5,325 → 633 tokens; $3.20 → $0.38), because a real flow includes governance tokens that are intentionally retained.

4. Where We're Going Next

4.1 DGM Maintenance Layer

The DGM (Digital Guild Master) maintenance layer is the next milestone. Every feature receives a per-feature Technician agent. On a configurable Monday schedule, the maintenance scheduler invokes each Technician through the opencode harness, runs the feature's maintenance routine, and reports a Seed-Status back to the Central YoloMoE Seeder — the same single write path used by all docking. A self-maintaining fleet whose maintenance work is itself governed and metered.

4.2 Custom opencode Tools

A set of custom tools on the opencode harness: 3-DGM gate check, zero-token ClawIntent emission, Claw execution against remote/GPU/local providers under a metered key, provider health, key minting and metering, settlement, seed reporting, estimate and trace, a DSYNC threat scan, and a first-time-use setup tool that captures the KPI Tokenizer prerequisites (identity, model in use, log source, KPI connection, emitter connection).

4.3 Cross-Platform DGM MCP Server

The same engine is exposed as a DGM MCP server (stdio and Streamable HTTP), so the identical tools work in Claude Code, Codex, Antigravity, VS Code, and any MCP-capable client. Each host supplies its own trigger: opencode native agents and scheduler, Claude Code headless runs on a cron, Antigravity's native cron sidecar, or a VS Code extension task.

4.4 DGM is Not Another MCP

DGM feature	MCP equivalent?	Why
Callable tool surface	Yes (only overlap)	Model-invoked, named, parameterized capabilities
Harness / connectivity	Yes	MCP is the harness standard; KPI logs harness IDs
ClawIntent contract	Partial	Structured invocation; signed, nonced, zero-budget
3-DGM Gate / DSYNC / policy	No	Governance; MCP has no gate or threat scan
Digital API Keys	No	Scopes, quota, expiry; outside MCP's scope
Central YoloMoE Seeder	No	Single write-path ledger; MCP has none
Token metering / settlement	No	Token economy; MCP has none
Technician + scheduler	No	Lifecycle; MCP has no scheduling
KPI progression tracking	No	Analytics; MCP has none

MCP answers "how does a model connect to tools, anywhere?" — a communication and integration standard. DGM answers "who is allowed, what does it cost, what happened, and who maintains the fleet?" — a governance, token-economy, and maintenance system. Orthogonal layers of equal standing. DGM rides MCP transport for reach; its substance lives above and beyond the protocol.

4.5 Launch

The opencode layer, the DGM MCP server, and the first-time setup flow ship as a standalone repository (deckergui/dgm-opencode) with npm packages @deckergui/opencode-tools and @deckergui/dgm-mcp, full documentation, and a Product Hunt launch.

5. Conclusion

DeckerGUI has progressed from a governance specification to a live, token-governed agentic ecosystem: an emitter console with a measured 88.11% token saving, a structural 99.4% reduction for docking agents, a progression-tracking KPI Tokenizer, and a public landing surface behind production tunneling. The next milestone — the DGM maintenance layer — extends the same governance to the maintenance of the fleet itself. DGM is not another MCP: it is the governance, economy, and maintenance layer that MCP deliberately leaves out — and the two work together as peers.