ARC Chain - Sharded AI Inference Across the Network

Block

Finalized

Validators

Blocks/sec

TPS

Chain

Total Transactions-

Accounts-

Peers-

Uptime-

Codebase101K LOC Rust

Tests1,234 passing

Regions6 continents

Nodes - 6 Continents

Sharded AI - One Model, Split Across the Network PIPELINE-PARALLEL

A model that cannot fit on any single node is split at transformer layer boundaries. Each request flows through the pipeline of shard-holders via HTTP, with BLAKE3 hashes verifying every hidden state in transit. Pure i64 arithmetic. 0% quality loss vs the source GGUF.

Model

Total Layers

Shards

Runs Served

Bytes Forwarded

If you tried this on 1 node ✗ OOM

- GB

Full model RAM. Most consumer/cloud nodes have 8 GB max. Won't load.

Sharded across the network ✓ FITS

- GB / node

Each node holds ~- layers. Together = full model. Loads everywhere.

Live Pipeline · Each box is one node holding a contiguous layer range Hover for details · Click ▶ to send a token through

No shards announced yet. Start nodes with --shard-start --shard-end

Type a prompt - watch the activation hop through every shard

Try:

Sends one token through all shards. Each hop verifies the previous shard's BLAKE3 hash. Last shard runs the LM head + argmax. Cached prompts return in ~100 ms (~200 tok/s).

Community Network 0 workers

0 jobs completed

Community nodes providing inference compute. Join with one command below.

🚀 Join the network as a shard holder in one command

Persistent service. Daily auto-update. Auto-detects your platform. macOS arm64/x86, Linux x86_64/aarch64. ~3 minutes from curl to running node.

curl -sSL https://raw.githubusercontent.com/FerrumVir/arc-chain/main/scripts/install-community-node.sh | bash

Or run the demo end-to-end without installing: curl -sSL https://raw.githubusercontent.com/FerrumVir/arc-chain/main/scripts/arc-demo.sh | bash

Or run the same prompt on every device in parallel (load balancing demo)

Parallel AI Inference (Load Balancing) LIVE

Each node runs a full copy of the SAME model. Same prompt → same hash on every device → cross-platform determinism proven.

Inference Nodes

Speedup

6.2×

Total Runs

Type a prompt - watch all 8 devices answer it together

Same prompt sent to all 8 nodes simultaneously. All produce identical hash = cross-device determinism proven.

Or click any device to run a quick test

Compares 1 node sequential vs 8 nodes parallel. Same prompt. Different machines. Identical hash.

ARC Chain TESTNET

Chain

Nodes - 6 Continents

Sharded AI - One Model, Split Across the Network PIPELINE-PARALLEL

Or run the same prompt on every device in parallel (load balancing demo)

Parallel AI Inference (Load Balancing) LIVE

Live Inference Feed

Recent Transactions