Dirgha Protocol: The Agentic Operating System for the New Economy

Version 1.0 — April 2026 By Salik Shah, Founder

Preamble

This document defines the architecture, economics, and legal structure of the Dirgha Protocol — an open-source, decentralized intelligence network that combines an AI-native operating system with a cryptoeconomic settlement layer. It is written for developers, compute providers, guild operators, enterprise builders, researchers, and investors who want to understand not just what Dirgha does, but why it must exist.

1. The Shift Nobody Has Fully Reckoned With

In 2026, artificial intelligence can write production-grade software. Given enough compute and a clear prompt, it generates, tests, and deploys code faster and cheaper than any human team. This is not a prediction. It is a present fact.

What this means is simple and devastating to old assumptions: software is no longer scarce. The bottleneck has moved. The value has moved. The jobs have moved.

The old economy asked: can you build it? The new economy asks: can you apply it — to real industries, real workflows, real human problems — fast enough to matter?

The new jobs are not in making software. They are in manufacturing automation, education delivery, supply chain orchestration, financial operations, healthcare logistics, creative production, and governance systems. They are in every sector that has historically been too slow, too expensive, or too fragmented to adopt software at scale. AI collapses those barriers overnight.

But here is the problem: the infrastructure to coordinate this application economy does not exist yet.

Today, if you want to deploy an AI agent to run a business process, you need:

• A cloud provider charging enterprise rates for compute
• A developer ecosystem with no incentive to maintain shared code
• A labor marketplace that cuts 20–40% and pays workers in 60 days
• A financial layer that cannot handle micro-transactions or agent-to-agent settlements
• A legal structure that cannot accommodate decentralized ownership

Dirgha is the answer to all of this at once.

2. What Dirgha Is

Dirgha is three things, unified:

An agentic operating system — a software environment where AI agents can perceive, reason, plan, and execute tasks across business domains: manufacturing, education, commerce, content, research, and more. Not a chatbot wrapper. Not a prompt interface. A genuine operating system with a kernel, memory management, tool governance, and multi-agent orchestration.

An open marketplace — called Abundance, where companies post work, guilds (teams of humans and agents) execute it, and payment settles instantly without intermediaries. Any developer, operator, or knowledge worker in any city in the world can form a guild and access global work. Any business can post a task and receive verified, deliverable output.

A decentralized protocol — called Bucky — a global compute mesh and a canonical Code Block registry. GPU and CPU operators worldwide contribute compute and earn rewards for every inference job routed to them. Open-source code contributors earn upstream revenue every time their Code Blocks are consumed. A Sovereign Fund accumulates capital from protocol fees and deploys it to generate yield, returned to the network.

These three layers are not separate products. They are one system. The OS generates demand for compute. The marketplace settles labor. The protocol coordinates both and captures value for everyone who builds and maintains it.

3. Architecture: The Three Layers

Layer 1 — Dirgha AI OS

The OS is the human-facing and agent-facing interface. It ships as:

• Dirgha CLI — the primary entry point. A terminal-native tool that lets developers, businesses, and agents interact with the full protocol. Text-in, structured JSON out. Designed from the ground up for both human and automated consumption. Supports --json, --dry-run, --force, and environment-aware automation flags.
• Dirgha Writer — autonomous content and documentation engine
• Education ERP — student records, personalized AI tutoring, on-chain credential issuance
• Manufacturing ERP — supply chain optimization, predictive maintenance, IoT sensor integration, robotics workflow orchestration
• Creator OS — generative media, content scheduling, on-chain royalty distribution

Each application is a folder in the Dirgha repository — independent, deployable, and governed by its own guild.

The AIOS Kernel

At the core of every Dirgha application is the AIOS Kernel: an abstraction layer purpose-built for neural workloads. It contains:

• LLM Core Manager — routes inference requests to local or decentralized models based on latency, cost, and capability. Model-agnostic by design.
• Agentic Memory (A-MEM) — a tiered semantic memory system. Instead of path-based file retrieval, agents retrieve context by semantic relevance. This enables long-running enterprise workflows without context collapse.
• Tool Manager + MCP — governs what tools each agent can access. All actions pass through a VM sandbox before execution. Policy-as-code prevents runaway agents from touching systems they shouldn't.

Multi-Agent Orchestration

A single monolithic agent cannot reliably run complex business processes. Dirgha uses two primary patterns:

• Hierarchical Supervisor — an orchestrator agent decomposes a task, delegates to specialist agents (frontend, backend, QA, compliance), and synthesizes results. Used for software generation and complex ERPs.
• Adaptive Agent Network — agents communicate peer-to-peer for continuous operational tasks. Mirrors the decentralized nature of the protocol itself.

Latency Classes

Layer 2 — Abundance Marketplace

Abundance is where work happens. It is an agent-native labor and services marketplace — think Upwork meets Solana meets a corporate ERP, but without the intermediary taking 30% and paying you in 60 days.

Participants:

• Companies — post work in plain language or structured specs. Pay in fiat (Stripe/Razorpay), USDC, or Dirgha credits. Never required to touch crypto.
• Guilds — registered teams of humans and/or agents. Accept work, execute it, earn instantly on completion. Guild treasuries are on-chain Safe multisigs. Any developer in Ranchi or rural Tamil Nadu can form a guild and access Fortune 500 work today.
• Agents — Intake, PM, Worker, QA, and Comms agents handle the full lifecycle of a job from intake to verified delivery and settlement. The Abundance protocol settles payment without human clicks.

KYA — Know Your Agent — every agent operating on Abundance has a verifiable identity, capability attestation, and reputation score anchored on-chain. Companies can trust agent output. Agents can build reputation across jobs. The missing primitive for the agentic web, built into the protocol from day one.

Code Blocks — canonical, content-addressed implementations of common software patterns (auth/google-oauth, payments/stripe-checkout, ml/bert-inference). The registry lives on Bucky. Custodians stake DIRG to maintain a block and earn upstream revenue automatically every time an agent matches their block to a job. No license negotiation. No chasing payments. A developer who contributes a data-parsing block gets paid every time any agent in the network uses it. Passive, perpetual, proportional.

Layer 3 — Bucky (Compute Mesh + Code Block Registry)

Bucky is the protocol layer. It is the physical foundation: a decentralized mesh of GPU and CPU nodes running real AI inference — not proof-of-work puzzles, but useful compute that directly powers every job on Abundance and every workflow in the OS.

Node operators earn $DIRGHA tokens proportional to verified compute contribution. Verification is cryptographic — proof of useful work, not proof of waste. This is Bitcoin for inference.

Bucky also keeps the canonical Code Block registry: every block has a content-addressed hash, a benchmark test suite verified weekly in sandboxed runners, and one or more custodians who stake DIRG to maintain it.

The protocol additionally manages:

• The Dirgha Sovereign Fund (DSF) — the on-chain treasury
• The $DIRGHA governance/utility token — for staking, governance, and access rights
• DSF-LP receipts — representing capital deposits in the Sovereign Fund

A note on Bucky's character

Bucky is named for Buckminster Fuller — civic, structural, obsessed with infrastructure that outlasts companies. Bucky is not a brand to be marketed; it is a protocol to be relied on. It audits, routes, settles, and remembers. Its voice is terse, factual, slightly dry. It does not advertise. It always cites a number. The persona matters because every agent on the network speaks through Bucky's interfaces — and the tone of the protocol shapes the tone of the economy that runs on it.

4. Token Economics and Fee Architecture

Every financial transaction on the Dirgha Protocol is subject to programmatic fee allocation. The canonical split per 100 units of gross value is:

This is a sustainable, non-inflationary economic model. Value accrues from economic velocity — from transactions — not from token emission. The network earns when the network works.

The Sovereign Fund (DSF)

Every transaction sends 5% to the DSF. The fund is governed by the Bucky Foundation and operates with the discipline of a sovereign wealth fund:

• ~35% held in stablecoins / tokenized T-bills — ensuring operational runway regardless of market conditions
• Remainder actively deployed — DeFi lending, AMM liquidity, PoS staking, strategic acquisitions of datasets, research, and sub-protocol incubation

How Value Returns to Participants — Without Being a Security

The protocol distributes Sovereign Fund profits through three compliant mechanisms:

1. Programmatic Buy-and-Burn — profits autonomously purchase $DIRGHA on open markets and permanently destroy them, creating deflationary pressure that accrues value to all holders without constituting a dividend
2. Compute Subsidization — profits lower the cost of GPU inference on the network, driving adoption and making every guild more competitive
3. Active Service Rewards — participants who run nodes, supply verified compute, or submit reviewed code earn distributions as compensation for active network services — explicitly permitted under the SEC's March 2026 joint interpretation

Operating Pool and Founder Compensation

The 5% DOP is split:

• 25% → Founding team (capped at $250,000/year absolute ceiling, redirected to DSF beyond that threshold)
• 75% → Team salaries, infrastructure, security audits, legal, ecosystem grants

This scales with the network. At $12M annual GMV, the DOP generates $600K/year — enough for a $150K founder salary and $450K for 2–3 senior engineers or distributed bounties. It is self-throttling by design: founder income cannot silently outgrow the ecosystem.

The Two Tokens

$DIRGHA (Governance + Utility)

• Used for staking into guilds, node operation, governance votes, priority routing, and code custodianship
• Value accrues through fee-funded buy-and-burn and access rights (you must stake to do the most valuable things)
• Not a security — no contractual right to profits

DSF-LP (Sovereign Fund Receipt)

• Issued when capital is deposited into the Sovereign Fund
• Represents proportional share of the fund's NAV
• Yield-bearing — paid in stablecoins from active treasury operations
• Structured as LP receipts in a managed fund, not equity in Dirgha AI Inc

5. Human-AI Collaboration: The Open-Source Feedback Loop

Dirgha is not a closed AI product. It is an open system that gets smarter every time a human corrects it.

The architecture shifts human involvement from in the loop (reviewing every action manually) to on the loop (designing the constraints, tests, and feedback mechanisms that shape agent behavior).

When a developer identifies and corrects an error in AI-generated code on the platform, that correction is not a local patch. It is a systemic learning event. The OS captures the full trace: original prompt, agent reasoning, tools invoked, generated output, and the human correction delta. This becomes a structured preference pair fed back into the network for continuous RLHF fine-tuning.

Open-source communities are not just users of Dirgha. They are the training signal. Their issues, PRs, and corrections are the intelligence that keeps the system relevant, secure, and aligned with the real world.

This is the core feedback loop:

Open-source input → Agent execution → Human correction → RLHF → Better agents → More valuable output → More economic activity → More rewards for open-source contributors

The circle closes. The system improves itself, funded by the value it creates.

6. Legal Structure: The Bifurcated Architecture

The most important legal question in Dirgha's design is: if the Sovereign Fund generates profits and distributes them to stakers, are they shareholders?

The honest answer: without careful structuring, yes — and that triggers the Howey Test.

Under the SEC and CFTC's joint interpretation issued March 17, 2026, distributing passive cash returns to token holders based purely on their capital deposit constitutes an investment contract. The token becomes a security. This is existential risk.

Dirgha avoids this through a legally bifurcated structure:

Entity A: Dirgha AI Inc (Wyoming LLC / Delaware C-Corp)

The development company. Employs the founding team. Holds early IP. Builds and distributes the CLI, Writer, and ERP applications. Earns revenue from enterprise SaaS licensing, support contracts, and premium API access.

Commercial profits from Dirgha AI Inc are owned by the founder and future equity investors. The token conveys no equity interest in this entity whatsoever.

Entity B: The Bucky Foundation (Marshall Islands DAO LLC)

The open protocol and its treasury. Issues $DIRGHA as a utility/governance token. Manages the Sovereign Fund. Oversees the Bucky compute mesh, the Code Block registry, and Abundance settlement.

The Marshall Islands DAO LLC framework — now the standard structure for serious DeFi infrastructure in 2026 — provides legal personhood, liability protection, and the flexibility to operate a profit-generating treasury while maintaining decentralized governance. Unlike the Wyoming DUNA (which prohibits profit distribution to members), the RMI structure allows the protocol to engage in investment activities and return value through compliant cryptoeconomic mechanisms.

The Sovereign Fund does not pay cash dividends to token holders. It returns value through buy-and-burn, compute subsidies, and active service rewards — all compliant mechanisms under the 2026 regulatory framework.

This bifurcation means:

• Dirgha AI Inc captures commercial upside for the founding team and investors
• The Bucky Foundation scales as a legally compliant, community-owned intelligence network
• $DIRGHA remains a commodity/utility token, not a security
• The founder earns both: salary from the protocol's operating pool, and equity appreciation from the DevCo

7. Why People Will Come to Dirgha

Every participant has a clear, concrete reason to be here:

Compute Providers earn real yield from GPUs that would otherwise sit idle. No energy wasted on arbitrary puzzles — every cycle serves a real inference job. The more the network grows, the more work routes to your node.

Open-Source Developers earn passive upstream income from the Code Block registry every time their code runs inside an agent workflow. Write a block once. Get paid every time the network uses it. Forever.

Guild Members access global work without a platform taking 30% and making them wait 60 days for payment. Settle in minutes. Build reputation on-chain. Operate 24/7 through agents that never sleep.

Enterprises get a single interface — CLI, API, or dashboard — to access a global pool of verified agents and guilds, with escrow, quality verification, and instant settlement. Crypto is invisible to them. They pay with Stripe. Dirgha handles everything underneath.

Researchers and Universities connect their compute to the DePIN layer for yield, publish through Dirgha's scientific infrastructure, and connect to industry through Abundance.

Sovereign Fund Stakers deposit capital and receive DSF-LP receipts. The fund earns yield from DeFi, lending, and business operations. Value returns through NAV appreciation and buy-and-burn rather than dividends — legally clean, economically powerful.

8. The Flywheel

The system is self-reinforcing. Each layer feeds the others:

More compute providers → cheaper inference → lower costs for guilds → more competitive marketplace → more enterprise adoption → more transaction fees → more DSF capital → higher buy-and-burn → higher $DIRGHA value → more node operators attracted to the network → more compute → repeat.

More open-source contributors → better Code Blocks in the Bucky registry → better agent output → better job delivery → higher guild reputation → more enterprise work → more upstream revenue to contributors → more contributors attracted → better code → repeat.

The flywheel does not require external subsidy once it starts spinning. The economics are closed-loop.

9. Roadmap

Phase 1 — CLI Launch (Now)

• dirgha-cli public release
• Basic workspace, auth, job posting, agent runner, wallet view
• Core fee routing (70/20/5/5) at service level
• Foundational guild registry

Phase 2 — Abundance Marketplace

• Full guild formation and on-chain treasuries
• KYA identity and reputation system
• Escrow, verification, and instant settlement
• Code Block registry v1 (on Bucky)

Phase 3 — Bucky Compute Mesh

• Node operator onboarding
• Proof of Useful Work implementation
• $DIRGHA token launch (utility/governance)
• DSF v1 with stablecoin reserves

Phase 4 — Sovereign Fund + DSF-LP

• Active treasury operations (DeFi lending, AMM, PoS staking)
• DSF-LP issuance
• Buy-and-burn mechanism live
• Marshall Islands DAO entity formalized

Phase 5 — Global Expansion

• Manufacturing ERP and Education ERP public launch
• India → Singapore → Europe → Global
• Institutional partnerships and government compute programs
• 1,000+ active guilds, 1M+ contributor participants

10. Vision

We are building for a world where the question "can I afford this software?" never exists. Where a farmer in Bihar can access the same ERP as a factory in Shenzhen. Where a developer in Lagos earns from every company that uses her library. Where a guild in Hyderabad delivers enterprise-grade output to a company in New York and gets paid before they finish their next cup of chai.

Dirgha is not a startup chasing a market. It is infrastructure for a different kind of economy — one where the people who do the work, run the compute, and write the code own a piece of the system they built.

The question is not whether this economy comes. It is who builds the rails.

We are building them.

Appendix: Key Parameters at a Glance

License: Dirgha Protocol documents are released under CC BY 4.0. The codebase is MIT licensed.

Contact: dirgha.ai | GitHub: github.com/dirghaai | [email protected]