CommonCompute
Log inGet started
Log in
Get startedDownload the Mac app
Open alpha·provider v1.7.14

A catalog of AI workloads.
Running on real Apple Silicon.

OCR, embeddings, background removal, text-to-speech — ten workloads live today, more shipping every release. Each one runs natively on the Apple Neural Engine or VideoToolbox, dispatched to a network of real, individually-owned Macs. One SDK call, a CLI, or an MCP server for your agent — priced before the job runs.

Start a job →View pricing⤓ Download Mac app
$ pip install commoncompute
quote locked before execution · Python + TS · signed receipts
◵ live routing model — job submitted → matched to a provider Mac → executed → returned
Live network · list prices
01OCR
$1.05
per 1k pages · 30% under AWS Textract · live
02Transcription
$0.252
per audio hour · 30% under OpenAI Whisper API · live
03Setup
1 call
vs days of IAM roles + job definitions
04Runtime
Apple Silicon
CoreML · Metal · Vision · native
§ 02What you can run

One catalog. Live today, shipping next.

Built for throughput, not realtime. If your workload can tolerate seconds of dispatch latency for a dramatic cost cut, it belongs here. Large embedding batches fan out across every available Mac and reassemble automatically — more throughput, still one locked price.

TaskRuntimeDescriptionUnit
01OCR / doc extraction
Apple Vision
Convert PDF archives, receipts, scanned contracts into structured JSON.
per 1k pages
02Transcription
whisper-large-v3
Turn call recordings, meetings, podcasts into searchable text — in bulk.
per audio hour
03Reranking
bge-reranker (MLX)
Sharpen RAG retrieval without the per-search markup of a rerank API.
per 1k queries
04iOS device testingSoon
XCTest · Simulator
Run your XCTest suite on real Apple Silicon instead of a device-cloud meter.
per test minute
05Translation
Apple Translation
Localize catalogs, docs, and subtitles at a fraction of per-character API rates.
per 1M chars
06Background removal
Vision · CoreML
Cut out product shots and portraits in bulk on the Neural Engine.
per image
07Embeddings
bge-small · apple-nl-300d
Index your corpus nightly. Swap in without changing a line of inference code.
per 1M tokens
08Video transcode
h264 · hevc · prores
Re-encode media libraries on Apple hardware encoders, not rented GPU time.
per output minute
§ 03Why Apple Silicon

The one hardware moat the hyperscalers can't rent you.

FIG.01
Frameworks clouds can't run
Apple Vision OCR, VideoToolbox encoders, AVFoundation, Apple Translation, and the Neural Engine only exist on Apple hardware. AWS, GCP, and Azure literally cannot offer them.
FIG.02
Silicon built for this work
Every M-series chip ships a dedicated ML engine and hardware media encoders. For OCR, transcription, translation, and transcode, that beats renting a training GPU by orders of magnitude on cost.
FIG.03
Hardware that's already paid for
The network runs on Macs that were already bought and already powered — homes, studios, mini farms. No data center, no cooling bill, no idle fleet to amortize into your price.
Already on OpenAI?Embeddings and transcription are drop-in — swap two env vars. Migration guide →
§ 04Two sides of the network

One marketplace. Bring jobs, or bring a Mac.

For developers

Drop-in replacement for the expensive parts of your AI stack.

Point your embeddings, Whisper, OCR, and rerank calls at Common Compute. Same interfaces you already use. A fraction of the cost. Deterministic quotes before every job runs — no more surprise month-end invoices.

embed.pypython 3.12
1  from commoncompute import Client
2  
3  cc = Client()
4  job = cc.embeddings.create(
5      model="bge-base",
6      input=open("corpus.jsonl"),
7  )
8  # → job.id, job.price, job.eta
Read the SDK docs →
Have a Mac?

Your Mac earns while it's idle.

Install the menu-bar app in minutes. It runs catalog workloads when your Mac is on power and your settings allow it, pauses the moment you need the machine back, and pays out weekly with an itemized earnings ledger.

menu bar · idle⌘,
Sharing — idle
Running catalog workloads
This week
$6.90
Tasks
214
CPU 14% · MEM 3.2 GB · THERM nominal
Install the Mac app →
§ 05Under the hood

Built for engineers who need to trust the result.

Six guarantees, spelled out. No asterisks, no "enterprise tier" gate.

Signed
Tamper-evident receipts
Every result ships with a signed receipt — cost, provider, input/output hashes. Platform-signed today; provider-side signing is next.
Gated
Scoped inputs
Every job input passes a trusted-download gate — HTTPS-only, private/loopback addresses blocked, size-capped, content-verified — before it runs.
Quoted
Deterministic pricing
No idle or queue charges, ever. Most workloads price by volume, quoted before you run — a few compute-bound jobs (training, builds, rendering) bill by measured run time.
Polyglot
Python + TypeScript
An OpenAI-shaped embeddings API — drop the official client in via our compat shim. Inline responses up to 256 inputs; larger batches return a job.
Cataloged
Status-gated workloads
Only live workloads are submittable. The API rejects a job for anything still coming soon or in preview, before you're ever charged.
Auditable
Open capacity API
Macs online, chip mix, per-workload availability — queryable through an unauthenticated public endpoint. No job-activity ledger yet.
§ 06Pricing

The price is locked before the job runs.

Every submission returns a quote first, capped by your max_spend_usd. You pay only for successful jobs — no idle charges, no hourly drift. Cheapest in the market on Apple-Silicon-native workloads.

TaskUnitTypical costvs. hyperscaler
OCR · Apple Visionper 1k pages$1.05−30% under AWS Textract
Transcription · Whisperper audio hour$0.25−30% under OpenAI Whisper API
Translationper 1M chars$14.00−30% under Google Translate
Embeddingsper 1M tokens$0.014—
See full pricing →list prices — workloads still rolling out are marked on the pricing page

Run one job. See it land.

The network's built from Macs like yours. Join as a developer or a provider.

Create an account →⤓ Download the Mac app
// $10 minimum to start · signed receipts on every job