Capture any text. Enrich it with local AI. Publish curated Streams. Everything stays on your machine — zero cloud, zero tokens, zero compromises.
All output below was generated by Qwen 3.5 running on-device. Relay never sent a single byte to a server.
In 2026, on-device AI is no longer a promise — it's a platform shift. Apple ships a 3B-parameter model running locally on iPhones by default. Google releases capable open-source models small enough for phones. The EU AI Act demands data sovereignty. The tools of thinking are moving from the cloud to the edge. Gearbox Relay is the transmission for this new reality.
Apple Foundation Models debut on-device. On-device AI becomes a first-class platform feature, not a research demo.
Gemma 4, Qwen 3.5 prove that capable models run at <200 MB. The cloud inference bottleneck breaks.
EU AI Act enforcement begins. Zero-click searches dominate the web. The privacy-first, local-first paradigm becomes the default expectation.
Gearbox Relay is the private operating system for your context. It captures what you notice, enriches it on your device, and lets you relay your curated perspective to others. No cloud ever touches your thoughts. No one can look inside.
Transmission is biological. Biology is mechanical.
Highlight, copy, screenshot. Your attention leaves a trace. Relay grabs it with source metadata — URL, title, author, timestamp.
On-device AI instantly tags, summarizes, and connects your capture to existing knowledge. No cloud, zero token cost.
Publish curated Streams. Others subscribe to your signal. Your private library stays sealed — you decide what to transmit.
Qwen 3.5 tags, summarizes, and connects every capture via llama-cpp-2. Embeddings via ONNX. All inference is local.
Full-text (FTS5) + semantic (LanceDB 384-dim vectors) hybrid search. Fallback to brute-force cosine on cold start.
Curate topic Streams. Each published Stream auto-updates. Subscribers see your curated feed in their own Relay gearbox.
AI-curated spaced repetition (SM-2). Daily review sessions surface thematic clusters from your knowledge graph.
All AI runs on your hardware. No per-token costs, no API keys, no cloud dependencies. Unlimited captures at no marginal cost.
The entire client — capture, enrich, search, publish, sync — is published under Apache 2.0. Build from source. Audit every line.
Markdown, JSON, ZIP. Your knowledge graph is yours — export it anytime with full metadata and vector indices.
AES-256-GCM end-to-end encrypted sync across devices. Argon2id key derivation. Zero-knowledge server stores only ciphertext.
*Creators keep 100% of subscriber revenue for the first 12 months. After that, a voluntary Transmission Fuel contribution (10–20%) keeps the relay running.
Weekly distillation of the most important papers and releases in on-device machine learning, with practical deployment notes.
Curated signal on encryption, data sovereignty, and practical privacy engineering. No theory — just shipping.
Official product updates, architecture deep-dives, and transparency reports from the Gearbox team.
Your personal transmission. Runs entirely on your machine.
TestFlight
Coming soonPlay Store Beta
Coming soon