From 2b826e07ad546daf5cb1079e251552249e024a70 Mon Sep 17 00:00:00 2001 From: Naveen Singh Date: Thu, 25 Dec 2025 12:19:33 +0530 Subject: [PATCH] Add Common Compute Network grant application --- applications/coco-ai-network.md | 284 ++++++++++++++++++++++++++++++++ 1 file changed, 284 insertions(+) create mode 100644 applications/coco-ai-network.md diff --git a/applications/coco-ai-network.md b/applications/coco-ai-network.md new file mode 100644 index 00000000000..4e031488aa6 --- /dev/null +++ b/applications/coco-ai-network.md @@ -0,0 +1,284 @@ +# Common Compute Network + +- **Team Name:** Common Compute Labs +- **Payment Details:** + - **DOT**: 12nNNM6RBnExmfvLy2cGAFF2Qp7HShXN2W6bh9jSN9XcjW6y + - **Payment**: 12nNNM6RBnExmfvLy2cGAFF2Qp7HShXN2W6bh9jSN9XcjW6y (USDC on AssetHub) + - **Preference**: 80% in vested DOT, 20% in USDC (AssetHub). We intend to keep the vested DOT staked in-network (e.g., nomination pools/validators) in accordance with program rules. +- **Level:** 2 + + +## Project Overview :page_facing_up: + +### Overview + +Common Compute Network is a decentralized infrastructure protocol that transforms edge devices into a distributed AI inference network, with economic incentives and value settlement built on Polkadot. Our project extends the existing Common Compute OS - a lightweight, AI-ready operating system currently deployed at scale - to create a tokenized network where compute providers earn rewards for serving AI inference requests. + +**Project Components:** +- **Common Compute OS**: Minimalist edge OS with pre-installed ai inference server +- **Substrate Runtime**: Custom pallets for compute resource management, task distribution, and token economics +- **Inference Protocol**: Decentralized protocol for routing AI requests to optimal compute nodes +- **Economic Layer**: Token-based incentive system for compute providers and consumers + + +The project integrates into the Polkadot ecosystem by: +- Utilizing Substrate for the core blockchain infrastructure +- Using the parachain primarily for transparency (device registry, job intents metadata), financial records, internal governance, and maintaining treasury +- Keeping AI inference off-chain via a gateway and device agents; on-chain is used for verifiable coordination and accounting +- Optionally enabling interoperability (e.g., XCM) later for cross-chain reporting and treasury flows, not direct dApp-facing inference + +We are building this because there is a clear technical gap and opportunity: +- Many AI workloads can execute on commodity edge hardware (especially quantized foundation open source LLMs), yet there is no standard way to discover, attest, and coordinate those resources. +- A Substrate-based registry and job-intent mechanism provides verifiable coordination, permissionless participation, and an audit trail for usage—well-aligned with blockchain primitives. +- Common Compute OS already standardizes the runtime on devices; adding a light client, agent, and minimal pallets enables a practical path from single-device inference to a networked, on-chain-coordinated system. +- Providing this capability natively within the Polkadot ecosystem unlocks AI features for parachains and dApps without depending on centralized inference providers. + +### Project Details + +**Architecture Overview:** + +``` ++------------------+ HTTP/REST +------------------+ +| AI Consumer | --------------------> | Gateway | +| (App/User) | <-------------------- | (API Bridge) | ++------------------+ +---------+--------+ + | + | Inference (off-chain) + v + +----------------------+ + | CoCo Device | + | (Edge Node) | +----------------------+ + + on-chain metadata/accounting attestation/heartbeat + +-----------------------------------------------+ + | Parachain | + | Transparency / Governance / Treasury | + +------------------------------+----------------+ + | + +---------v----------+ + | Polkadot Relay | + | Chain | + +--------------------+ +``` + +**Core Components to be Built:** + +1. **Substrate Runtime Pallets (non-dApp-facing):** + - `pallet-device-registry`: Manages compute node registration, capabilities, status, and last heartbeat + - `pallet-compute-jobs`: Records job intents and acceptance metadata for auditability (no inference on-chain) + - `pallet-treasury-governance` (scaffold/scope for later): Treasury records, internal governance hooks, payout events + - `pallet-reputation` (later): Tracks compute node performance and reliability for transparency + +2. **Compute Node Integration (off-chain inference):** + - Go-based edge agent on Common Compute OS with Substrate light client integration for attestations/accounting + - CoCo Edge Node runtime wrapper and orchestration in Go (requests do not go on-chain) + - Resource monitoring and reporting via Go services; posts minimal metadata on-chain for transparency + - Secure inference execution environment with existing CoCo Edge Node + Go agent coordination + +3. **Inference Protocol:** + - Load balancing and optimal node selection + - Task validation and verification mechanisms + - Result aggregation for improved reliability + - Privacy-preserving inference options + +4. **Gateway (scope for Level 2):** + - Minimal REST gateway for AI requests; writes/verifies on-chain metadata for coordination and accounting + - Basic client examples and docs (SDKs deferred to a follow-up) + +**Technology Stack:** +- **Blockchain**: Substrate framework with custom pallets +- **Compute Layer**: Go-based services and orchestration (leveraging existing api-server) +- **Edge Runtime**: Go agent interfacing with Substrate + CoCo Edge Node runtime +- **AI Framework**: CoCo Edge Node with support for multiple model formats +- **Bridge Layer**: Go-Substrate integration via gsrpc +- **Networking**: libp2p for peer-to-peer communication + Go HTTP services +- **Frontend**: React-based dashboard for network monitoring (existing uniphy.commoncompute.org) +- **APIs**: REST endpoints via enhanced Go API server + +**Data Models:** + +*Go Service Layer:* +```go +// Minimal Go service data structures +type ComputeNode struct { + NodeID string `json:"node_id"` + Capabilities []ModelCapability `json:"capabilities"` + ResourceStats ResourceMetrics `json:"resource_stats"` + CoCoEdgeNode string `json:"coco_edge_node"` + Status NodeStatus `json:"status"` +} + +type InferenceRequest struct { + ModelID string `json:"model_id"` + Prompt string `json:"prompt"` + Strategy InferenceStrategy `json:"strategy"` + MaxTokens int `json:"max_tokens"` + Temperature float32 `json:"temperature"` +} +``` + +Note: The device-produced usage record (JSON + sr25519 signature) is specified and delivered under Milestone 2. + + + +### Ecosystem Fit +- Role: infrastructure-facing pattern for coordinating off-chain AI inference on Polkadot; not dApp-facing +- Users: AI/ML developers, node operators, infra teams; later, parachains for reporting/treasury via XCM +- Need: transparent device discovery, job intents, and verifiable usage artifacts without on-chain inference +- Differentiators: Substrate-native coordination; edge-first OS/agent; privacy-preserving local execution with a live OS baseline + +## Team :busts_in_silhouette: + +### Team members +- Naveen Singh - Creator ( System Architecture, Product Design) +- Rachit Sharma - Contributor ( Backend, DevOps, Tooling) +- Stephen Pryce - Contributor ( BizDev, Marketing, Finance) +- Devaj Mody - AI/RL Engineer (Reinforcement Learning, AI Systems) + + +### Contact +- **Contact Name:** Naveen Singh +- **Contact Email:** io.naveens@gmail.com +- **Website:** https://commoncompute.org + +### Legal Structure +- **Registered Address:** Not Incorporated ( Tentative Location - Thailand/Singapore/UAE ) +- **Registered Legal Entity:** Tentative: UniPhy Foundation + +### Team's experience + +**Creators Experience:** + +- **For Profit Startups**: + Built an enterprise SaaS Servify, going public at $2.3B in NSE India + Building PhyFarm (Raising - $5M Series A) - currently in growth stage scaling across India to serve 300Mn Farming and rural population + +- **Impact & Sustainability**: Regenerative initiatives at RezenLabs - aligning technology with environmental and social impact rezenlabs.com +- **Edge Computing Expertise**: Successfully developed and deployed 10,000+ IoT devices with active user base across multiple hardware platforms +- **Systems Architecture**: Deep experience with distributed systems, performance optimization, and production-scale deployments with over 500Mn Userbase on cumulative projects done so far + + +**Previous Web3 Foundation Grants:** None (first application) + +### Team Code Repos +- https://github.com/w0w/common-os +- https://github.com/w0w +- https://github.com/DevajMody + + + +## Development Status :open_book: + +**Current Implementation:** +- **Common Compute OS v1.0**: Production-ready with automated setup +- **CoCo Edge Node Integration**: Fully functional API for local LLM inference +- **Network Configuration**: mDNS, hotspot fallback, mobile-optimized web UI +- **Performance Benchmarks**: Tested on various Raspberry Pi configurations +- **Documentation**: Comprehensive setup and usage guides + +**Deployment readiness:** +- **Hardware pipeline**: 100 device pre-orders; distributor pipeline forecasting up to 10,000 units within 12 months. +- **Fleet ops**: Reproducible OS image, first-boot automation, and ARM64 agent packaging for batch flashing and staged rollouts. + +**Research and Development:** +- Analysis of token economics for sustainable compute networks +- Evaluation of Substrate pallet architecture for compute marketplaces +- Performance testing of different consensus mechanisms for task distribution +- Privacy-preserving inference techniques compatible with blockchain verification + +**Community Engagement:** +- Building active dev community around Common Compute OS for Real world uses cases +- Feedback from early adopters on edge AI deployment challenges +- Collaboration discussions with potential parachain integrators + +**Technical Foundation:** +The project builds upon solid technical foundations: +- Proven OS deployment and management system +- Established AI inference pipeline with CoCo Edge Node +- Understanding of edge device limitations and optimization strategies +- Clear vision for blockchain integration architecture + +## Development Roadmap :nut_and_bolt: + +This roadmap is scoped for a Level 2 grant with two concrete milestones that deliver working code, tests, and documentation, building on the existing Common Compute OS. + +### Overview +- **Total Estimated Duration:** 10–12 weeks +- **Full-Time Equivalent (FTE):** 2 +- **Total Costs:** 30,000 USD (proposed) +- **DOT %:** 80% + +### Milestone 1 — Device Registry Pallet, Agent, and CLI + +- **Estimated Duration:** 5–6 weeks +- **FTE:** 2 +- **Costs:** 13,500 USD (45% of total) + +| Number | Deliverable | Specification | +| -----: | ----------- | ------------- | +| **0a.** | License | MIT for agent/CLI; Apache 2.0 for pallets (or project standard) | +| **0b.** | Documentation | Guide to run a local Substrate node, build/install pallet, register a Common OS device from ARM64, and query state | +| **0c.** | Testing and Testing Guide | Unit tests for pallet (>70% coverage on extrinsics/weights); integration test for device registration and status updates | +| **0d.** | Docker | Dockerfile/docker-compose for local node + example device agent | +| 1. | `pallet-device-registry` | Substrate pallet implementing: `register_device`, `update_metadata`, `set_status`; Events, Errors, Weight annotations; storage for owner, capabilities (model, RAM, arch), status, last_heartbeat | +| 2. | Device Agent (ARM64) | Service for Common OS that generates/loads sr25519 key, posts heartbeats/status, updates capabilities; systemd service template and ARM64 build | +| 3. | `coco-cli` | CLI to register/update devices and query registry (works on macOS/Linux and ARM64) | +| 4. | Demo | Scripted demo: launch local node → register device from a Pi (or ARM64 container) → update status → query via RPC | + +### Milestone 2 — Job Intents, Gateway Prototype, and Metering Record + +- **Estimated Duration:** 5–6 weeks +- **FTE:** 2 +- **Costs:** 16,500 USD (55% of total) + +| Number | Deliverable | Specification | +| -----: | ----------- | ------------- | +|| **0a.** | Documentation | End-to-end tutorial: submit job intent → device acceptance → local CoCo Edge Node execution → produce signed usage record | +|| **0b.** | Testing and Testing Guide | Pallet unit tests; gateway/agent integration test running against local node; mock/fake CoCo Edge Node optional for CI | +| 1. | `pallet-compute-jobs` | Minimal job intents pallet: `submit_job_intent`, `accept_job`, job lifecycle states; storage for requester, model, max_price, SLA fields; events/errors/weights | +|| 2. | Gateway Prototype | Minimal REST bridge (subset of OpenAI/CoCo semantics) that posts on-chain job intents and listens for acceptances; forwards job details to device agent | +|| 3. | Device Agent Extension | Job runner invoking local CoCo Edge Node with provided prompt/model; streams logs locally; emits signed usage record (duration, approx tokens/bytes) | +| 4. | Usage Record Spec | JSON + signature format (sr25519) for device-produced metering record to be used in later settlement work | +| 5. | Demo | Scripted E2E demo on local devnet: API request → on-chain intent → device acceptance → job execution → usage record artifact | + +Notes +- All code will include clear build/run instructions and avoid leaking secrets in logs or examples. + +## Future Plans +- Follow-up: settlement/escrow and payments; smoldot/light client integration +- Technical: expanded model support; enterprise features (private networks, SLA); privacy-preserving options +- Community: workshops and infrastructure contributions across Middle East, South & East Asia + +## Referral Program (optional) :moneybag: + +- **Referrer:** N/A +- **Payment Address:** N/A + +## Additional Information :heavy_plus_sign: + +**How did you hear about the Grants Program?** Web3 Foundation Website and community research + +**Additional Information:** + +- DOT stewardship commitment: We commit to keep 100% of the vested DOT in the Polkadot ecosystem (staked via nomination pools/validators). Staking rewards will be used to offset infra/CI, security hardening, and maintenance releases. We will not market-sell the vested DOT. +- Community participation: We are pursuing long-term participation in growing the Blockchain community and network in the Middle East, South & East Asia (Thailand, Malaysia, Singapore, India, UAE) - Our bet is on Polkadot/Jam network as highest performing L1 chain in market and best suitable for real world use cases we are looking to build on CoCo + +**Market Validation:** +- Growing demand for AI features in Web3 applications +- Cost advantages over centralized AI services (estimated 70-90% cost reduction) +- Privacy first design +- Edge AI usage growing across industries + +**Technical Innovation:** +- There is no widely adopted pattern in Polkadot for verifiable coordination of off-chain AI inference. This work proposes a minimal, reusable template (pallets + agents) other teams can adopt. +- Novel token economics for compute resource allocation +- Integration of edge computing with blockchain consensus +- Privacy-preserving inference with blockchain verification + + +**Strategic Importance:** +This work introduces a reusable pattern for Polkadot networks that need transparent coordination of off-chain AI inference: on-chain device discovery, job intents metadata, verifiable usage artifacts, and governance/treasury support. It reduces dependency on centralized inference APIs while keeping computation off-chain, and offers a pallets + agents template other teams can adopt. + +**Design Principles:** + +AI inference benefits from decentralization when coordination and usage are verifiable, and when data can be processed close to the source. Common Compute Network implements these principles in a practical way: a permissionless registry of edge devices, on-chain job intents, and an agent that executes workloads locally via the CoCo Edge Node. This proposal delivers the minimal building blocks needed to demonstrate the model end-to-end on Polkadot.