# DIMAGGI AI

> Six independent products from DIMAGGI AI, from agent controls to funding preparation and verified work records.

## About
DIMAGGI AI builds independent products. Tool Guard governs agent tool calls at runtime. d4research keeps technical investigations intact when the coding agent changes. Tenwa works on autonomous network operations. Caliber tests reasoning and judgment. Funding Readiness Kit prepares business records for an application. NewWork keeps proof and payment tied to a completed job.

## Author
Maggie Nanyonga, Founder & CEO
https://www.linkedin.com/in/margaret-nanyonga/

## Website
https://dimaggi.ai

## Key Pages
- [Home](https://dimaggi.ai/): Tool Guard — policy decision engine (Core) and MCP enforcement gateway (Enterprise) for AI agent tool calls
- [Products](https://dimaggi.ai/products): Six independent products, not a bundled platform
- [d4research](https://dimaggi.ai/products/d4research): Local-first research with one authoritative thread and local Memo context
- [Funding Readiness Kit](https://dimaggi.ai/products/funding-readiness-kit): Sort business records, find missing documents, and prepare an application packet
- [NewWork](https://dimaggi.ai/products/newwork): Keep assignments, proof, verification, and payment in one work record
- [d4research on GitHub](https://github.com/dimaggi-ai/d4research): The open-source early-access workspace and generic orchestration engine
- [Resources](https://dimaggi.ai/resources): Tool Guard Core documentation — getting started, architecture, policy authoring, integration, operations
- [Insights](https://dimaggi.ai/insights): Articles on AI agent governance, runtime policy decisions, and audit
- [Tool Guard Core on GitHub](https://github.com/dimaggi-ai/tool-guard-core): the open-source engine — policy evaluation, classifiers, hash-chained audit, the tg-proxy /evaluate decision service, CLI
- [Contact](https://dimaggi.ai/contact): Partner with us

## Documentation
- [Quick start](https://dimaggi.ai/resources/getting-started)
- [Architecture](https://dimaggi.ai/resources/architecture)
- [Core vs Enterprise](https://dimaggi.ai/resources/oss-vs-enterprise)
- [Policy schema (YAML)](https://dimaggi.ai/resources/creating-policies)
- [Agent integration](https://dimaggi.ai/resources/integration)
- [Production deployment](https://dimaggi.ai/resources/operating)

## Articles
- [The Maintainer Loop: Keeping Books on the Bookkeeper](https://dimaggi.ai/insights/ontology-debt-maintainer-loop): An LLM reviewer once invented a paper that nearly landed in my related-work table, and a dead API key once produced an audit that looked immaculate because nothing had been asked. Both burns became permanent regression tests, then an architecture: a maintainer loop that keeps my benchmark current, where the agent proposes, deterministic code decides, and a human merges. Nineteen bugs before the loop's first run, one more after it, and every gate built so the agent never needs to be trusted.
- [A Technical Framework for Frontier AI Capability Evaluations](https://dimaggi.ai/insights/ai-model-containment-solutions): Frontier AI models evaluated with reduced safety refusals exhibit multi-step reasoning, zero-day discovery, and persistent sandbox escape. Three disclosed 2026 incidents (the OpenAI ExploitGym breach of Hugging Face, NanoGPT PR #287, and Claude Mythos Preview) share one root cause: the boundary was enforced by policy, not hardware. This framework covers Firecracker microVM air-gapping, out-of-band eBPF kernel tracing, deceptive canary infrastructure, and process-based reward models.
- [Ontology Debt: A Consistency Ledger for LLM World-Models](https://dimaggi.ai/insights/ontology-debt-consistency-ledger): A model a standard accuracy score would call 96 percent correct insisted a row of untouched dominoes had fallen, then had not, then had. Ontology Debt is a small open-source tool that catches exactly this. You declare the invariants your LLM must hold, it probes any Anthropic or OpenAI model with deterministic scoring, and it keeps two numbers it refuses to blend, violations against your commitment and the model contradicting itself, on a debt ledger that remembers until they are paid down.
- [Guardrails Are Not Instructions](https://dimaggi.ai/insights/guardrails-are-not-instructions): An instruction in a system prompt is a suggestion the model is statistically likely to honor, not a rule it must obey, and you cannot fix that with more instructions. The only guardrail that holds is one the model does not get to reason about. DIMAGGI Tool Guard moves the decision outside the agent, into a deterministic engine that returns allow, deny, escalate, or flag on every tool call and writes a tamper-evident record of why.
- [Measuring Judgment in an AI-Assisted World — Caliber](https://dimaggi.ai/insights/measuring-judgment-in-an-ai-assisted-world-caliber): AI made fluency free, so the scarce asset is now judgment that holds up. Why the signal that told you who could actually think went dead — and how to measure it again.
- [Ground Truth Is the Moat](https://dimaggi.ai/insights/ground-truth-is-the-moat): The real competitive advantage in autonomous network infrastructure is not the sensors or the models, which anyone can buy, but the verified corpus of physical truth and the loop that produces it. Competitors can accumulate it. They cannot purchase it. And the corpus you have not started is the one you cannot buy later.
- [The Infrastructure Schism](https://dimaggi.ai/insights/infrastructure-schism): Every governance framework being built for autonomous infrastructure controls what an agent may do, not whether what it believes about the physical world is true. The model is provably wrong, and at the physical layer the action does not reverse. This paper names the Infrastructure Schism, the unmonitored gap between the trusted model and physical reality, and argues that verified ground truth, not more permissions, is the precondition for handing irreversible control to machines.
- [The Sovereign NOC: A Hybrid Human-Agent Operations Case Study](https://dimaggi.ai/insights/sovereign-noc-hybrid-ai-agents-network-operations): This case study applies the hybrid human-agent squad blueprint to network operations, showing how NOCs can coordinate human engineers and AI agents through shared OKRs, authority tiers, runtime policy firewalls, structured I-PASS handoffs, audit trails, self-interrogation protocols, and a phased pilot roadmap for governed autonomy.
- [Building Hybrid Human-Agent Squads with Shared OKRs](https://dimaggi.ai/insights/hybrid-human-ai-agent-teams-operating-model): A practical enterprise blueprint for designing and governing hybrid human-AI agent teams. The paper covers shared OKRs, role design, AI agent accountability, risk-tiered autonomy, the Workslop Tax, handoff contracts, audit trails, and an 8-week pilot model anchored in NIST AI RMF principles.
- [The Network as the AI Grid: Edge Inference, Silicon Economics, and Sovereign Infrastructure for the Last Mile](https://dimaggi.ai/insights/network-as-ai-grid-edge-inference-sovereignty): In the generative era, the digital divide is shifting from a data transit problem to an intelligence access bottleneck. By re-architecting distributed telecom footprints into sovereign inference grids, emerging economies can escape data colonial structures and deliver resilient, last-mile utility. This paper reviews vendor lock-in dynamics across legacy systems, provides a comprehensive Total Cost of Ownership (TCO) breakdown, and explores edge-native resilience frameworks.
- [Enforcing Runtime Policy: The Architecture of Trust in Clinical AI](https://dimaggi.ai/insights/clinical-ai-runtime-policy-accountability): Capability is not safety. In high-stakes clinical environments, bridging the gap between raw LLM reasoning and patient safety requires an execution-layer policy firewall. Here is why static governance falls short and how runtime interception establishes verifiable trust.
- [AI Compute Economics: Why Real Output Is Growing 18x Faster Than the Spending](https://dimaggi.ai/insights/ai-compute-economics-measurement-gap): Nominal AI spending grew 144 percent a year. Real output grew 2,600 percent. That gap decides who captures the surplus and who gets crushed.
- [The Two Workloads. Why Training-Shaped Infrastructure Is Now Stranded Capital](https://dimaggi.ai/insights/two-workloads-stranded-capital): Inference is now the majority of hyperscaler AI compute, with analyst estimates ranging from 60 to 70 percent. The provisioning ratios that defined training-era infrastructure are wrong for a workload they were never built to serve.
- [Audit-After-The-Fact Is Bankrupt. Governance Now Has a Latency Budget](https://dimaggi.ai/insights/governance-has-a-latency-budget): Agents act fifty times faster than humans, but the policy layer most enterprises bought was designed for ticketed review. Either policy moves to the tool boundary, or it stops working.
- [World Models and Zero‑Shot Planning: The Imagination Engine for Physical AI](https://dimaggi.ai/insights/world-models-and-zero-shot-planning): How predictive, action-conditioned simulation enables grounded reasoning and planning—complementing language models with a deeper sense of reality.
- [Frontier AI Models Have a Blind Spot for Ambiguity](https://dimaggi.ai/insights/complexity-preservation-ai-benchmark): New benchmark results show where social cognition breaks—and it's not where you think. Complexity Preservation, the ability to hold ambiguity without collapsing to a verdict, is the weakest skill across all capable models.
- [Compute With Legal Identity](https://dimaggi.ai/insights/compute-with-legal-identity): Infrastructure 'now' carries jurisdiction, residency, and operational autonomy as physical properties. Architects still designing for the borderless cloud are building legacy tech.
- [The Hardware Bet You Can't Actually Audit](https://dimaggi.ai/insights/hardware-bet-you-cant-audit-silent-ai-failure): At the scale of modern AI clusters, hardware failure isn't dramatic. It's silent. And most CTOs are approving infrastructure without knowing how to detect it.
- [The Quiet Crisis: Why Moving Fast on AI Is Breaking the Pipeline That Builds Your Next Leaders](https://dimaggi.ai/insights/quiet-crisis-ai-leadership-pipeline): The first real AI labor shock won't be the layoffs. It will be the juniors who never got the chance to become seniors.
- [Hybrid Pricing for AI Support in B2B SaaS](https://dimaggi.ai/insights/hybrid-pricing-ai-support-b2b-saas): Most B2B SaaS companies will get AI support pricing wrong unless they combine a subscription base, an outcome-tied variable layer, and a premium trust tier. Here's how Intercom, Zendesk, Sierra, and Salesforce are actually pricing it, what's working, and what fell apart.
- [The Resolution Ledger: The Missing Product Behind Outcome-Based Pricing](https://dimaggi.ai/insights/resolution-ledger-outcome-pricing): Outcome pricing sounds simple until a customer asks why you billed them for 3,214 resolutions last month. A resolution ledger is the boring, spreadsheet-shaped product that makes outcome-based AI support pricing auditable and defensible.
- [From Pilot Purgatory to the Orchestrated Era: A Manifesto](https://dimaggi.ai/insights/from-pilot-purgatory-to-the-orchestrated-era-a-manifesto): Most companies aren’t failing at AI. They’re succeeding at the wrong version of it — and the gap is about to become irreversible.
- [The Sovereign Enterprise: Rewiring for the Agentic Era](https://dimaggi.ai/insights/the-sovereign-enterprise-rewiring-for-the-agentic-era): Most organizations are currently stuck in "Pilot Purgatory," deploying chatbots that act as expensive bandaids on legacy wounds. To win, you must stop treating AI as a guest in your house and start treating it as the foundation. This requires a fundamental rewiring of your data, your governance, and your people.
- [The AI Workload Hierarchy: Why Global AI Infrastructure Is a Portfolio Problem, Not a Capacity Problem](https://dimaggi.ai/insights/the-ai-workload-hierarchy-why-global-ai-infrastructure-is-a-portfolio-problem-no): Power, cooling, latency, and geography are redrawing the capital allocation map. The window to get positioning right is narrowing fast.
- [The Architecture of the Forkable Firm: Why the Future of Leadership is 'Org Code' and Vertical Co-Design](https://dimaggi.ai/insights/future-of-leadership-agentic-organizations-vertical-codesign): The model now dictates the org chart, not the other way around. Conway's Law has reversed, and most firms aren't ready.
- [Dirty Data to Clarity: How AI Transforms Raw Data into Insight](https://dimaggi.ai/insights/dirty-data-to-clarity): Transforming raw randomness into structured intelligence. Leveraging AI in data governance.

## Topics
AI governance, AI security, agentic organizations, runtime policy enforcement, AI audit trails, Conway's Law, role transformation, LLM workflow automation, policy-driven AI, vertical co-design