System Overview and Benefits
This page covers what EdgeCoder is, what it ships today, and why it matters.
What EdgeCoder Is
EdgeCoder is a decentralized mesh network where every device contributes compute to run AI models at the edge. There is no central server. Every node is a unified agent -- running coordinator, worker, and inference in a single process. Nodes discover each other via a gossip mesh and distribute tasks based on capability and availability.
No cloud accounts. No API keys to a hosted service. Every participating device is both a provider and a consumer of AI compute.
Shipped Features
Unified Agent
Every node runs coordinator + worker + inference in one process. There is no separate service deployment, no multi-container orchestration, and no control plane to stand up. A single binary joins the mesh and is immediately productive.
Chat-First Desktop App
A Tauri-based IDE ships with a Monaco code editor, streaming chat interface, conversation history, model management, and a live mesh visualization panel. This is the primary user surface for interacting with the network.
Gossip Mesh
HTTP-based peer discovery and task broadcasting. Nodes announce themselves and relay peer lists to build a full mesh view. Task distribution uses a claim-delay protocol to prevent duplicate execution across nodes.
Local AI Inference
Every node runs model inference locally via Ollama. iOS nodes use llama.cpp for on-device inference. No inference request ever leaves the executing node's hardware.
BLE Local Mesh
Bluetooth Low Energy enables offline peer-to-peer task routing between nearby devices. This provides a connectivity layer that works without any network infrastructure -- useful for field deployments, air-gapped environments, and mobile-to-mobile coordination.
Credit Economy
Nodes earn credits by contributing compute and spend credits to submit tasks. The ledger is an append-only ordering chain anchored to Bitcoin, with Lightning Network settlement for real-time payouts.
External IDE Support
VS Code, Cursor, and Windsurf connect to any EdgeCoder node through an OpenAI-compatible HTTP endpoint. Developers keep their existing editor workflow and route completions through the local mesh instead of a cloud API.
Security
Ed25519 request signing authenticates every inter-node message. Nonce-based replay prevention, per-node rate limiting, and a blacklist audit chain provide defense in depth. Envelope encryption for task payloads is staged for a future release.
Benefits
No cloud dependency
All inference runs locally on commodity hardware. There is no hosted backend, no metered API, and no third-party data processor in the loop.
Permissionless mesh
Any device can join the network and begin contributing compute or submitting tasks. There is no enrollment approval gate and no vendor lock-in.
Cryptographically auditable
Every credit transaction is recorded in an append-only ordering chain anchored to Bitcoin. The full history is independently verifiable by any participant.
Privacy-preserving
Task data stays on the executing node. Prompts, code context, and model outputs are never transmitted to a central service or stored outside the device that ran the inference.
When EdgeCoder Is a Strong Fit
- Teams with private code or compliance constraints that prohibit sending source to cloud APIs
- Organizations that want AI-assisted development without a recurring cloud spend line item
- Operators building internal or hybrid compute networks from existing hardware
- Mobile and field teams that need AI inference without reliable internet connectivity