Eigen Labs launches project Darkbloom to route AI inference through idle Apple Silicon Macs

Eigen Labs has unveiled Project Darkbloom, a research initiative that routes AI inference requests through idle Mac computers instead of traditional data centers. The project is live in research preview and claims it can cut inference costs by roughly half compared with major aggregators, while giving node operators 95% of revenue.

How Project Darkbloom works

Darkbloom matches inference requests with verified Mac nodes through a coordinator system. Developers use an OpenAI-compatible API, while Mac owners run a hardened provider agent that processes requests locally.

The project is designed to address the trust problem of running prompts on another party’s device. Eigen Labs says the provider process blocks debugger attachment and external memory inspection, uses binary integrity checks to verify software matches network expectations, and relies on Apple’s Secure Enclave for hardware-backed attestation. The system also includes recurring challenge-response checks to confirm nodes maintain expected security states.

Eigen Labs also emphasizes that the coordinator remains a trusted component, rather than relying on “decentralized” messaging to obscure that dependency.

Cost model and supported workloads

Darkbloom’s economics are based on removing several layers of centralized inference costs. Traditional stacks include hyperscaler margins, API provider fees, facility overhead, cooling, and networking. Eigen Labs’ approach shifts the model toward marginal electricity costs, since the hardware is described as already owned and powered by participating operators.

The project currently supports text generation, image processing, and speech-to-text workloads. Eigen Labs’ announcement frames the 95% revenue share to operators as a key incentive to participate, though it does not provide production-load benchmark details in the available material.

Engineering challenges highlighted by the team

Project lead Gajesh Naik says the hardest engineering problems were not request routing, but the surrounding security and reliability requirements. These include code signing, release consistency, attestation timing, model lifecycle management, and handling disconnects and corrupted files.

“When binary hashes are part of the security model, release engineering becomes security engineering,” the team noted in their announcement. Cold starts, memory pressure, and network failures aren’t edge cases in a distributed system. They’re Tuesday.

What is available now

The research preview includes the full stack: the coordinator, a hardened provider agent, Secure Enclave integration, operator tooling, and a web console. Eigen Labs says the codebase is open-sourced and that a technical paper has been published.

Context within DePIN

Darkbloom is positioned within the broader DePIN (decentralized physical infrastructure) trend, which has gained momentum over the past year. Similar efforts such as Render, Akash, and io.net have explored decentralized compute for AI workloads, particularly GPU-based services. Darkbloom’s focus on Apple Silicon aims to carve out a different niche by using consumer hardware for inference.

No token has been announced. For now, it remains a research project testing whether idle laptops can supplement—or eventually compete with—the data center infrastructure that dominates AI investment.