How layr-labs/d-inference works
Darkbloom is a decentralized private-inference network: consumers submit OpenAI-compatible requests, the coordinator authenticates, routes and bills them, and provider Macs run inference in-process before streaming results back. The repository also includes a user console and an administrative UI for provider, account, network and payment operations. The README documents security and attestation paths; the sampled source excerpts are partial, so unverified internal wiring is intentionally omitted.
Main entry points: api Coordinator API · ProviderLoop+InferenceHandler.swift Request handler · dispatch.go Inference dispatch · MultiModelBatchSchedulerEngine.swift Model scheduler
Components 21
Consumer console
- User console
console-ui/src/app - Chat interface
console-ui/src/app/chat/page.tsx - Provider dashboard
console-ui/src/app/providers - Network explorer
console-ui/src/app/stats - Earnings and billing UI
console-ui/src/app/billing
Coordinator services
- Coordinator API
coordinator/api - Inference dispatch
coordinator/api/dispatch.go - Provider attestation
coordinator/api/app_attest.go - Billing and payouts
coordinator/api/billing_handlers.go - Base rewards
coordinator/payments/baserewards/engine.go - Network statistics
coordinator/api/network_series.go
Provider runtime
- Request handler
provider-swift/Sources/ProviderCore/ProviderLoop+InferenceHandler.swift - Model scheduler
provider-swift/Sources/ProviderCore/Inference/Engine/Scheduler/MultiModelBatchSchedulerEngine.swift - Engine bridge
provider-swift/Sources/ProviderCore/Inference/Engine/Bridge/EngineV2Bridge+Submission.swift - Vision prefill
provider-swift/Sources/ProviderCore/Inference/Vision/EngineV2VisionPrefill.swift - Capacity quoting
provider-swift/Sources/ProviderCore/Coordinator/CapacityQuoteEngine.swift - MLX inference
provider-swift/Sources/ProviderCore/Inference/Engine
Network operations
- Admin UI
admin-ui/src/app
Other components
- Consumer / SDK
- Apple Silicon GPU
- Stripe
Connections 22
- Consumer / SDK → Coordinator API: submits request
- User console → Coordinator API: calls APIs
- Chat interface → Coordinator API: sends chat
- Coordinator API → Inference dispatch: routes inference
- Inference dispatch → Request handler: dispatches encrypted request
- Request handler → Model scheduler: admits request
- Model scheduler → Engine bridge: submits tokens
- Engine bridge → MLX inference: runs model
- MLX inference → Apple Silicon GPU: executes inference
- Request handler → Inference dispatch: streams encrypted frames
- Inference dispatch → Consumer / SDK: streams response
- Coordinator API → Provider attestation: verifies providers
- Provider dashboard → Coordinator API: manages provider
- Admin UI → Coordinator API: queries operations
- Coordinator API → Billing and payouts: handles billing
- Billing and payouts → Stripe: processes payouts
- Coordinator API → Base rewards: settles rewards
- Network explorer → Network statistics: requests statistics
- Earnings and billing UI → Coordinator API: requests account data
- Coordinator API → Network statistics: serves network data
- Model scheduler → Vision prefill: prepares media
- Request handler → Capacity quoting: quotes capacity