Sidecar ML vs PromptSpend

Turn your iPhone into a private, on-device AI server for your whole network. Sidecar ML runs a lightweight HTTP server on your iPhone that exposes Apple's built-in machine-learning models to any device on your local network. Tap Start Server, scan the QR code, and your Mac, PC, Raspberry Pi, or browser can call powerful on-device AI over plain REST — no cloud, no accounts, no API keys, no usage fees, and no data leaving your network. Everything runs on the phone. Nothing is uploaded to a server and there's nothing to pay for. (The one exception, music recognition, matches against Apple's catalog and tells you so up front.) ONE IPHONE, A DOZEN AI CAPABILITIES • Chat with the on-device large language model through a chat-completions endpoint that's compatible with the popular OpenAI API format — streaming, JSON mode, and structured output included. Point existing tools and SDKs at your phone and they just work. • Generate images from a text prompt in a choice of art styles. • OCR — pull text out of photos, receipts, and screenshots with per-line boxes and multi-language support. • Read barcodes and QR codes (EAN, Code128, and more). • Classify images across roughly a thousand categories, or produce image embeddings for similarity search. • Remove backgrounds and isolate subjects or people as clean PNG cut-outs. • Detect faces and facial landmarks, body pose, and hand pose. • Scan documents with automatic edge detection and perspective correction. • Convert speech to text and text to speech, with a wide range of voices (including Personal Voice). • Translate text between languages completely offline. • Analyse text: language, sentiment, named entities, tokens, and embeddings. • Classify sounds across hundreds of categories, or identify the music that's playing. BUILT FOR DEVELOPERS AND MAKERS • Clean REST + Server-Sent Events API — call it from curl, Python, JavaScript, or any HTTP client. • Chat endpoint compatible with the OpenAI format, so many existing tools and SDKs work unchanged. • Automatic discovery on your network via Bonjour. • Honest capability reporting: ask the server what it can do and it tells you exactly what's available on your device, and why. • Optional bearer-token authentication to lock the server down on shared networks. FREE WEB CONSOLE AND CODE EXAMPLES A ready-to-use web console — plus Python examples, a command-line client, and full API documentation — is free and open-source. Download the single-file console and browse the source here: https://github.com/AndersFB/Sidecar-ML-Console/ Open the console in any browser, enter your phone's address, and try every capability with a friendly UI — nothing to install. GOOD TO KNOW • The on-device large language model and image generation require an Apple Intelligence-capable iPhone. The other capabilities run on a much wider range of devices. • The server runs while the app is in the foreground; a keep-awake option helps during longer sessions. • Some features (offline translation language pairs and certain speech models) download their models on first use. Turn the iPhone in your pocket into a private AI endpoint for your projects, your home lab, and your prototypes — Sidecar ML.

Know the tab before you build. PromptSpend turns a representative AI conversation into an understandable cost forecast before the first production invoice arrives. Paste sample text or enter known token counts, choose your traffic, and see the cost per conversation, day, month, and year. COMPARE THE SAME WORKLOAD Select up to four LLMs and compare their estimated costs side by side. PromptSpend calculates pasted text separately for each model family, ranks the results, and shows the monthly difference from the lowest-cost option. SEE THE FULL COST STORY Explore input, output, prompt-cache writes, long-context tiers, reasoning assumptions, batch discounts, and conversation growth. The Cost Brief, sensitivity lab, and savings playbook help explain what changes the number and which levers may reduce it. CURRENT PRICES WITH EVIDENCE PromptSpend downloads a validated public catalog, shows when sources were checked, flags disagreements for review, and refuses to hide an expired offline cache behind a reassuring number. Open the evidence and public sync status whenever you want to inspect the work. OPTIONAL PRICE ALERTS Create and manage an email alert for price changes without creating an account. Your email address and alert choices are sent only when you opt in, and every alert includes a way to unsubscribe. Native push notifications are not part of this release. SHARE THE COST STORY Create a readable estimate, four-model comparison, CSV export, restorable website scenario, or Estimate Receipt. Shared results contain derived counts, assumptions, costs, and source dates—not your pasted prompt text. AUDIT AN EXISTING CONVERSATION PromptSpend Receipt provides visible instructions to send in your AI conversation. Import the assistant's structured result, review its estimates and limitations, and share an image or readable text. Imported results are processed locally and cleared when you leave the Receipt screen. LEARN AS YOU GO Short guides explain tokenization, context growth, caching, pricing provenance, and the limits of any estimate. Global Search and a guided tour make deep features easy to find. MAKE IT YOURS Choose system, light, or dark appearance; four accessible accent colors; and cool or warm canvas styles. The Market Pulse ticker can be paused and respects Reduce Motion. PRIVATE BY DESIGN Pasted text is processed on your device and is not uploaded or saved by PromptSpend. There is no native account, advertising SDK, behavioral analytics, cross-app tracking, or tracking identifier. Saved scenarios and appearance settings stay on your device. Optional alerts transmit only the email address and choices you submit for that feature. PromptSpend estimates published API charges from the workload and assumptions you provide. Separately billed cache-storage time is disclosed but excluded unless a retention duration is known. Actual bills, model quality, latency, availability, regional pricing, negotiated rates, and provider behavior can vary. Validate a representative workload before making a production decision.