OnePlus 8T Mobile Lab System Map

简体中文 · English

The phone is not a bag of unrelated tools. It is one recoverable, composable, and verifiable mobile experiment runtime.

System map of the OnePlus 8T mobile lab, covering platform engineering, real-device automation, authorized security, and AI

View or download the original 1536×1024 artwork.

The current artwork uses Chinese labels. This page carries the complete architecture in English so the information remains searchable and accessible without reading text embedded in the image.

How to read the map

Repository boundaries

Repository Owns Does not own
oneplus-8t-mobile-lab Hub, teaching material, device baseline, integration labs, and public evidence Private configuration, credentials, firmware, or large backups
pocket-pentest Methods for authorized mobile security, CTF, networking, and hardware labs Unapproved targets or instructions that bypass scope controls
android-device-test ADB/UI automation, stability, performance, and reproducible device evidence Treating one screenshot as complete acceptance evidence
android-ai-stack OpenCode, Claude Code, Happy, model routing, on-device inference, plus the Mobile Lab Control Skills, Android source, and release APK Publishing real tokens, endpoint URLs, local provider profiles, or signing keys
xiaohei-phone-agent User-facing assistant product: invocation, short voice sessions, risk policy, notification drafts, and observable Android actions Advertising CPU KWS as DSP, background message sending, or requiring ordinary users to install root/pentest environments
happy-relay-deploy Optional remote relay, TLS, and Tailnet deployment A mandatory dependency for the phone-local workflow

Six useful compositions

Composition What it enables Boundary to preserve
AI + testing Draft cases, drive a real device, localize failures, and structure reports Model suggestions never replace assertions and device evidence
AI + security Plan authorized labs, invoke tools, interpret output, and write postmortems A person confirms scope, targets, and stop conditions
Assistant + AI runtime Xiaohei uses deterministic commands first and selects a model profile only for bounded complex tasks The model suggests a plan; Xiaohei still owns local policy, confirmation, and final action
Local + cloud Use a small model for offline triage, then route difficult work to a cloud model Keep model selection separate from service lifecycle; never publish credentials
Phone + remote Stay self-contained on the phone, then connect a workstation or relay only when needed Default to loopback; remote exposure needs explicit authentication and encryption
Labs + Wiki Compile real cases into guides, failure records, and cross-linked knowledge Separate facts, inferences, and unverified hypotheses

Mobile Lab Control separates the phone control surface into three independent widgets: Happy service, llama.cpp service, and model profiles. Each widget shows its own current state; service buttons manage lifecycle, while model buttons only change configuration. The app has no Internet permission, does not read tokens, endpoints, or the CC Switch database, and can invoke only fixed allow-listed actions through its Termux bridge. Browse the Android source and build guide or download a release-signed APK from GitHub Releases.

xiaohei-phone-agent is the product action layer above this runtime. Manual, Android Assistant, foreground CPU KWS, and validated device DSP entries converge on one short-command contract before local policy selects an Intent, notification, or visible Accessibility action. It can consume adapter-exposed model profiles without starting services on profile change; neither pocket-pentest nor a remote relay is an ordinary-user prerequisite.

Capability is not residency

This is a system architecture map, not evidence that every service is active at once. A hardened resting state keeps unused Happy, llama.cpp, Frida, and debug endpoints stopped. The tiny local model is a low-cost offline fallback, not a claim that the phone can efficiently host desktop-class models. Security work remains limited to owned systems, training targets, or explicitly authorized scope.

Continue with the device baseline, AI toolchain integration lab, or project map.