Remote Pi vs Peak Pocket AI

Remote Pi turns your iPhone into the control surface for the AI coding agents running on every machine you work from. Pair your phone once by scanning a QR code on your Mac. From then on your devices form a mesh: the agents on your machines talk to each other, and your phone is the trusted key that authorizes new peers and lets you step in from anywhere. WHAT YOU CAN DO • Scan a QR code to pair with a machine in seconds • See every active agent session across all your paired machines in one list • Follow the conversation live as your agent streams its work • Approve or reject tool calls with a tap — you stay in control of what runs • Move between machines without losing context YOUR PHONE IS JUST THE AUTHENTICATOR You don't run the heavy work on your phone. The agents live on your computers; your phone holds the cryptographic identity that vouches for new devices and gives you a window into what's happening. Pairing uses Ed25519 device keys over TLS. WORKS WITH THE HARNESS YOU USE Remote Pi rides alongside the coding-agent setup you already run on your machines instead of replacing it. OPEN SOURCE & SELF-HOSTABLE Remote Pi is open source. Start with the public relay, or point the app at your own self-hosted relay so the connection between your devices runs on infrastructure you control. Built for developers who want their agents within reach — on the couch, on the move, or away from the desk.

Peak Pocket AI is a lab for language models that run on your iPhone. Chat privately with open models, then look under the hood: how fast they generate, how confident they are in every word, how well they remember a long prompt, and how your phone holds up under sustained load. PRIVATE BY DESIGN Models run on the device. Your chats, notebook, photos, and settings never leave your iPhone. No account, no analytics, no ads. Once a model is downloaded, everything works in airplane mode. A LAB NOTEBOOK FOR EVERY GENERATION Every reply is recorded with its seed, settings, timings, and device state. Rerun any run with identical settings and see whether the output reproduces, character for character. Export runs as CSV or JSON, or as a signed report that anyone can check for changes. LIVE TELEMETRY Watch tokens per second, time to first token, memory, and thermal state as a reply streams in. SEE WHAT THE MODEL WAS THINKING Confidence shading shows how sure the model was of each word. Tap any word to see the alternatives it nearly chose, with their probabilities. Reasoning models show their thinking above the answer. EVALUATIONS Score models on built-in question sets for knowledge, math, tool calling, and structured JSON output, or import your own. Compare two models on the same questions with confidence intervals. EXPERIMENTS Run the same prompt and seed across models or settings. Measure how speed and memory change as the context fills, and test long-context recall: hide a fact at different depths of a long prompt and chart how often the model finds it. ENDURANCE AND ENERGY Measure what sustained generation does to your phone: throughput over time, throttling, thermal state, and battery used. An energy-per-token estimate is shown when a run is measured on battery. VISION AND APPLE'S ON-DEVICE MODEL Ask vision models about a photo, processed on the device. On iPhones with Apple Intelligence, chat with and evaluate Apple's built-in model alongside open models. BRING YOUR OWN MODELS Browse Hugging Face for GGUF models with a fit estimate for your phone's memory, including gated models with your access token. Optionally connect an OpenAI-compatible provider with your own API key; remote chats are clearly marked. MAKE IT YOURS Four themes in light and dark, including High Contrast, a colorblind-safe confidence scale, science personas, and share cards for any run or comparison. Built by Peak Summit Labs.