Compute marketplace · Early access
Apple Silicon compute. Dedicated AI inference.
Choose a supported model and find an approved Apple Silicon Mac for 30 days of managed text inference. One customer per machine, with no per-token charge. Or create a farmer account and supply an eligible Mac.
Already registered? Renter dashboard/Farmer dashboard
30 days
Published dedicated rental term
1 customer
Per dedicated machine rental
MLX
Model serving on Apple Silicon
Terms and runtime: provider documentation. Paid rentals accrue the provider share approved for that node. Free test allocations earn nothing.
Check supply before you plan a rental.
Choose a model, check compatible live machines and review the effective quote. Tier defaults do not reserve capacity. Check the current stock response below before funding a rental.
The decision in detail Current product reference
Three model configurations. One dedicated Mac per rental.
The reviewed catalog supports Qwen3.8 27B at 4-bit, 8-bit and BF16 (16-bit). Choose by model requirements and available machines, then evaluate your workload.
- Per paid rental term
- 30 days
- Per allocated machine
- 1 customer
- Chat completions and streaming
- Text
Qwen3.8 27B · required memory
Zero-based scale. Each metric uses its own unit and scale.
View all catalog values as a table
| Format | RAM (GB) | Disk (GB, approx.) | Context (tokens) |
|---|---|---|---|
| 4-bit | 24 | 16 | 16,384 |
| 8-bit | 48 | 30 | 32,768 |
| BF16 (16-bit) | 96 | 55 | 32,768 |
01 / For renters
Run a catalog model
through the API.
Your application sends model requests. The provider runs the serving process on a dedicated Mac. Choose the model and test the workload that matters to you.
Choose a catalog model
Compare the published 4-bit, 8-bit and BF16 (16-bit) configurations. Match memory and context to the work you need done.
Explore the catalogKnow what you rent
The documented rental supplies a model API for 30 days. Shell access and a remote macOS desktop are outside its scope.
Read the rental guideTest before you depend on it
Measure task quality, first-token latency and simultaneous requests. Decide how your application handles an unavailable node.
Plan your evaluation02 / For providers
Your Mac supplies compute.
Your farmer account manages it.
Start with an account at farmer.proxies.sx. It is required to connect a provider node and gives you access to compute management, earnings and payout requests.
Start with a Mac you already own
Check the current model requirements before buying hardware. Paid rentals accrue your approved share; listing and free tests earn nothing. Accrual, settlement and completed payouts are separate states.
Calculate revenue and operating costsHave an account? Open your compute dashboard.
- 01
Create and verify your farmer account
An account is required to provide compute. Sign up at farmer.proxies.sx and verify your email before opening protected portal sections.
Create an account - 02
Complete the agreement and get your key
Sign the Partner Agreement, then open Spawn API key, select Compute supplier and create a named key. Copy it once and save it privately.
Spawn API key - 03
Connect an eligible Mac
Inspect an existing installation first. Check memory, disk and native Node/Python, then use the current agent guide. Enter the supply key only in the hidden terminal prompt.
Installation guide - 04
Complete verification and supplier approval
Keep the measured model ready. Admission needs a passing benchmark within 24 hours, a countersigned agreement and explicit admin approval. Review the approved price and provider share, then enable List for rent in your farmer Compute page.
Compute dashboard
03 / The compute library
Guides for renting and providing compute.
Current product guides with practical checks, source references and clear operating limits.
Choose your inference setup
For rentersRent a Mac for Dedicated AI Inference
Evaluate a dedicated Apple Silicon inference rental: 30-day terms, supported models, API limits, stock and the checks to make before paying.
LLM Model Catalog: Memory & Context Limits
Compare Qwen3.8 27B model IDs, pinned revisions, quantization, RAM, disk and context limits. Check the supported catalog against current rental stock.
Dedicated AI Inference Pricing & Costs
Compare approved Mac rental quotes and current tier defaults. Understand 30-day balance payments, manual renewal, request limits and useful-output costs.
4-Bit vs 8-Bit LLM Inference: Memory & Quality
Compare 4-bit, 8-bit and bf16 LLM inference on Apple Silicon. Evaluate model quality, weight memory, context, runtime settings and cost per accepted result.
Dedicated LLM Inference Privacy Guide
Review prompt retention, supplier access and request handling for dedicated Mac inference. Check the published relay policy, application logs and RAG data.
Mac Inference vs GPU Cloud: Rental Guide
Compare managed Mac inference, bare-metal Macs and GPU cloud by API access, CUDA needs, billing term, model control and operating responsibilities.
Put your Mac to work
For providersRent Out Your Mac: Provider Setup Guide
Create your required farmer account, verify email and install the Mac compute agent. Follow benchmark and supplier approval through listing and earnings.
Apple Silicon LLM Inference: Memory Guide
Check Mac memory for LLM inference. Understand weights, KV cache, memory pressure, disk space and the current 24 GB, 48 GB and 96 GB catalog requirements.
Mac Compute Earnings & Cost Calculator
Estimate Mac provider earnings with your approved rental price and share. Compare occupancy, electricity and costs; separate accrual, settlement and payouts.
MLX Compute Agent Troubleshooting
Troubleshoot the current Mac compute installer, pinned Python runtime, model downloads, launchd, approval and offline status with read-only checks.
Build around the endpoint
For developersCompute API for AI Agents: Integration Guide
Integrate dedicated Mac inference with scoped API keys. Review request limits, streaming, unsupported options, errors and safe discovery before checkout.
RAG Inference on Apple Silicon: Setup Guide
Plan RAG with dedicated Apple Silicon inference. Separate retrieval and generation, budget context, test source support and enforce document permissions.
LLM Inference Benchmarks on Apple Silicon
Measure Apple Silicon inference with TTFT, completion latency, throughput, concurrency and task quality. Download a worksheet and define repeatable tests.
04 / For agents and developers
Discover the catalog.
Build around the model API.
Read models, published tiers and stock without credentials. Inference requires a renter account and an active rental. Keep orchestration, retrieval and external tools in your application.
GET /v1/peer/compute/catalog
Models → memory requirements → context limits
Match the catalog to current marketplace stock.
Proxy MCP and x402 integrations are documented separately. They do not establish compute purchasing support. Proxy access for agents.
Before you rent or list a Mac
- What does dedicated compute provide?
- One customer's model requests use the rented Mac for a published 30-day term. Your application calls an inference API. For Xcode builds, remote desktop or arbitrary server software, compare a service with operating-system access. Compare access and workloads.
- How much memory does a provider need?
- The supported catalog requires 24 GB for its 4-bit model, 48 GB for 8-bit and 96 GB for BF16 (16-bit). The current installer requires at least 26 GB free disk for the smallest setup. Benchmark and supplier approval determine admission. Check Mac requirements.
- Does the API support every agent feature?
- The customer documentation supports chat completions and streaming, with a 4096-token output cap and a 60-request/minute limit keyed by client IP and endpoint handler. Additional key/account limits may apply; one job runs per node. It rejects tools, stop sequences and response_format fields. Check that contract before choosing the endpoint for your application. Review the API limits and key scopes.
- What happens to prompts and responses?
- The published policy allows temporary relay storage. Completed-job content is normally scrubbed after one hour; job records expire after 24 hours, subject to cleanup timing. Suppliers administer their own Macs. Read the privacy evidence and limits.
Current product sources: provider requirements, model catalog, customer documentation. The dashboards require sign-in. Documentation describes supported behavior; benchmark the selected machine before relying on it.
Choose your starting point
Find inference for your application.
Or bring a Mac to the network.
Published by PROXIES.SX. The product reference records supported behavior. Current inventory, quotes and supplier terms come from the linked APIs and portals. Early Access has no availability SLA or earnings guarantee.