Decentralized compute · settled in Bitcoin

A global market for AI compute, priced in sats.

EdgeCoder turns idle machines into a coding swarm. Agents sell measured seconds of CPU and GPU time, requesters buy them with Lightning, and every hardware class trades against Bitcoin on an open order book. Your code runs on your machines first; only what you allow crosses to approved public capacity.

How it works

Measured work, market price, Lightning settlement.

No tokens to speculate on. One credit is one server-measured second of work on a known hardware class, and its sats value is locked the moment it is earned.

Contribute or buy

Install the agent to sell idle capacity, or buy credits with a Lightning invoice at the day's market rate.

Match on the book

Agents post an ask per hardware class, tasks carry a bid. A task is claimed only when the ask clears the bid, and it trades at the ask.

Verify and pay

The coordinator measures run time itself, samples results for redundant verification, and credits the worker at the agreed price.

Withdraw in sats

Earned sats are paid to your Lightning wallet on request or automatically to a Lightning address. Infrastructure hosts are paid from the network fee pool.

Pricing

Every hardware class has its own pair, priced in sats.

Each class is quoted in sats per unit-hour, converted from public market rates for compute and Bitcoin, recalibrated daily with outlier rejection and a daily move cap, then discovered on the order book.

btcUsd = median(Coinbase, Kraken, Bitstamp, mempool.space)
satsPerUnitHour = usdPerHour(class) / btcUsd × 1e8 × (1 + markup)
satsPerCredit = satsPerUnitHour / (creditsPerSecond × 3600)
× model size (≤3B 1.0 … >35B 4.0) × modality (vision, thinking) × scarcity
trade = agent ask ≤ task bid → priced at the ask, locked on the ledger
ClassHardwareUnit-hour
cpu.stdCommodity cores, < 32 GB30.8 sats/h≈ $0.02/h
cpu.hi≥ 32 cores or ≥ 128 GB123 sats/h≈ $0.08/h
apple.mApple silicon, < 32 GB unified memory185 sats/h≈ $0.12/h
apple.m.hiApple silicon, ≥ 32 GB unified memory463 sats/h≈ $0.30/h
gpu.c8Consumer GPU, ≤ 8 GB VRAM154 sats/h≈ $0.10/h
gpu.c16Consumer GPU, ≤ 16 GB VRAM308 sats/h≈ $0.20/h
gpu.c24Consumer GPU, ≤ 24 GB VRAM (RTX 4090 class)540 sats/h≈ $0.35/h
gpu.dc40Datacenter GPU, ≤ 48 GB1,079 sats/h≈ $0.70/h
gpu.dc80Datacenter GPU, > 48 GB (H100 class)2,776 sats/h≈ $1.80/h
Run infrastructure

Host the network, earn the fees.

A share of every settled task and every purchase goes to a fee pool that is distributed to infrastructure hosts by uptime and by what they run. Hosting a Bitcoin node for the network is paid work.

Weight
1.0×

Coordinator

  • Schedules tasks, runs the order book
  • Anchors ledger checkpoints
  • Serves local inference for chat
Weight
+1.0×

Bitcoin node

  • Bitcoin Core or Knots
  • Anchoring and fee estimation for the mesh
  • Block data for wallets and peers
Weight
+1.0×

Lightning node

  • LND or Core Lightning
  • Settles purchases and payouts
  • Earns routing fees on top
Built for teams

Private by policy, auditable by design.

Private policy boundaries

Decide what stays on your machines, what may cross into public capacity, and which nodes are eligible to take it.

Signed everything

Agents and coordinators hold Ed25519 keys. Requests, results, and ledger records are signed; task payloads are encrypted end to end.

Anchored ledger

A hash-chained ledger of claims, results and earnings, with periodic OP_RETURN checkpoints on Bitcoin for third-party verification.

Sandboxed execution

Remote code runs in Docker with no network, read-only filesystem and resource limits. No sandbox, no swarm work.

One mesh, many devices

macOS, Linux, Windows, iOS, Docker and VS Code agents; coordinators federate and gossip tasks across regions.

Operational access control

Passkeys, SSO, node enrollment tokens, operator approval and admin tooling for managed deployments.

Start with an account or an agent.

Create an account to buy compute and manage your wallet, or install the agent and start earning sats for idle time.