GridOne AI

Don't just use AI. Help build it.

Comment

Inspiration

Artificial intelligence is becoming one of the most transformative technologies, yet the infrastructure required to build and run it is increasingly concentrated on "gifted" players. At the same time, billions of personal devices contain useful capacity that often sits idle.

GridOne AI explores a more participatory model: not AI built only for humanity, but AI infrastructure built with humanity (a democrat player).

Don't just use AI. Help build it.

What it does

GridOne AI is a working prototype of a collaborative compute network. People become Builders by opting in their idle device capacity through Smart Contribution controls.

The coordinator divides a supported job into bounded, allow-listed microtasks and assigns them to available workers. Results remain untrusted until task-specific checks and redundant execution verify them. Only verified work earns prototype AI Credits, and every credit links to an auditable receipt.

The Build Week prototype demonstrates:

  • two independent worker processes;
  • Smart Contribution and manual pause controls;
  • task leasing, timeout, retry, and reassignment;
  • deterministic verification;
  • an intentionally tampered result that is rejected and safely requeued;
  • a live Builder dashboard with task state, audit events, and credit receipts.

How we built it

The prototype uses Node.js 20 built-ins for the HTTP coordinator, scheduler, worker processes, verification, ledger, and tests. The dashboard uses browser-native HTML, CSS, and JavaScript. It has no third-party runtime dependencies and requires no API key.

Workers accept only versioned, allow-listed task types—not arbitrary remote code. The demonstrated workload creates deterministic feature-hash vectors from public or synthetic text. The verifier checks digest, dimensions, token counts, and numerical output before the append-only prototype ledger issues credits.

How we used Codex and GPT-5.6

Codex with GPT-5.6 was our continuous product and engineering collaborator. It helped us pressure-test the original concept, separate the long-term cooperative vision from an honest one-week MVP, define trust boundaries and non-goals, implement the coordinator and worker protocol, build the dashboard, write the verification suite, execute the multi-worker demo, deploy the public build, and keep this submission aligned with what judges can actually run.

The key human decisions were the cooperative positioning, avoiding arbitrary remote code, choosing an independently verifiable workload, and refusing to overclaim privacy, global scale, or economics.

Codex Session ID (/feedback): 019f76be-0e27-7f80-be0d-8f827aaf8469

Challenges we ran into

  • Finding AI workloads that are both divisible and independently verifiable.
  • Treating unknown workers as untrusted without making redundancy erase the economic value.
  • Preserving Builder control while scheduling useful work.
  • Explaining an ambitious vision while clearly separating real, simulated, and future features.
  • Avoiding the false assumption that task splitting alone guarantees privacy.

Accomplishments that we're proud of

  • A complete contribution-to-verification-to-credit loop.
  • A visible rejection and recovery path for invalid work.
  • Five passing tests, including a real multi-process end-to-end run.
  • A reproducible public demo and a desktop proof using separate operating-system processes.
  • Transparent product boundaries: prototype credits are simulated and non-redeemable.

What we learned

Distributed compute is not automatically cheaper, greener, private, or trustworthy. Verification is part of both the product and the economics. The strongest experience is not a hardware dashboard—it is purposeful participation with visible control.

GridOne is designed as a complementary community-powered layer for suitable workloads, not a replacement for hyperscale data centers.

What's next

We plan to pilot with 25–100 opt-in Builders, measure useful versus redundant work and energy, add stronger isolation and reputation, test a second workload class, create a developer quickstart, explore sustainable AI Credit partnerships, and investigate cooperative governance so Builders have a voice in the network they power.

Try it

For the desktop proof, clone the repository, use Node.js 20+, run npm run demo, and open http://127.0.0.1:8787. Run npm test for the verification suite.

Built With

Share this project:

Updates