Ponsbloom
Private AI,
in bloom on idle Macs
Ponsbloom connects over 100 million Apple Silicon machines sitting idle directly to AI demand. Prompts are encrypted end-to-end, every answer is signed by hardware you can verify. OpenAI-compatible. ~50% cheaper than centralized APIs — operators keep 95% of revenue.
Three layers of markup, removed
AI compute today flows through GPU makers → hyperscalers → API providers, each taking a cut. Meanwhile 100M+ consumer Macs idle ~18 hours a day. Ponsbloom collapses the supply chain to a single hop — encrypted demand in, signed answers out.
Trust, without access
Operators run your inference — they cannot see your data. Every path from prompt to answer that an operator could observe has been eliminated.
Encrypted end-to-end
Requests are encrypted on your device. The coordinator routes ciphertext — only the target node's hardware-bound key can decrypt it.
Hardware-verified
Every node holds a key generated inside Apple's tamper-resistant Secure Enclave. Attestation traces back to Apple's root certificate authority.
Hardened runtime
Inference is locked at the OS level: debugger attachment blocked, memory inspection blocked, no shell access. Nothing to extract from a live process.
Traceable to hardware
Every response is signed by the exact machine that produced it. The full attestation chain is published for independent verification.
OpenAI-compatible API
Point your existing OpenAI SDK at Ponsbloom and go. Streaming, function calling, speech-to-text — same shape, lower cost, hardware-verified.
from openai import OpenAI client = OpenAI( base_url="https://api.ponsbloom.ai/v1", api_key="<your-api-key>", ) stream = client.chat.completions.create( model="gemma-4-26b", messages=[{"role": "user", "content": "Explain zero-knowledge proofs"}], stream=True, ) for chunk in stream: if chunk.choices[0].delta.content: print(chunk.choices[0].delta.content, end="", flush=True)
Per-token. No subscriptions.
Idle hardware has near-zero marginal cost — savings pass straight through to you.
| Model | Input / 1M | Output / 1M | Savings |
|---|---|---|---|
Gemma 4 26B Fast multimodal MoE — 4B active | $0.042$0.084 | $0.22$0.44 | 50% |
Qwen 3.5 27B Dense frontier reasoning | $0.10$0.20 | $0.78$1.56 | 50% |
Qwen 3.5 122B MoE 10B active — best quality | $0.13$0.26 | $1.04$2.08 | 50% |
Nemotron 3.5 Lightning NVIDIA — low latency | $0.065$0.13 | $0.18$0.36 | 50% |
GPT-OSS 20B OpenAI open weights | $0.02$0.04 | $0.10$0.20 | 50% |
Speech-to-Text Cohere Transcribe · per audio minute | $0.0010$0.0020 | — | 50% |
Your Mac earns while you sleep
Install the agent, keep your Mac plugged in, collect USD. You keep 95% of revenue — Ponsbloom takes no platform fee — roughly 90% profit margin after electricity.
Install the agent
One command. The agent handles enrollment, attestation, and model serving on macOS.
Stay plugged in
Scheduling is built in — serve only when your machine is idle, on power, or on your own hours.
Get paid in USD
Earnings accrue per token served. Track everything live in the operator console.
Curated, attested, ready
Start in under a minute
Sign in with email, get an API key, make your first private call — or plug your Mac into the network and start earning tonight.