Smartphone App Intelligent Group Control
20-¥ Self-Built Group-Control Router + LLM Task Orchestration · Carry Device Pairings on the Go
A multi-device collaborative control system that combines LLM task orchestration with a 20-¥ self-built ESP32 ADB group-control router. The local proxy / pairing router carries all multi-device pairing info on the go, so a fleet of phones can be assembled into an ad-hoc group-control workbench on any network. An LLM converts natural-language instructions into executable click / input sequences; the entire run is visible on three tracks — screen stream + Chain of Thought + step timeline — with prev / next / follow-latest navigation.
Core Modules
Control Console · Chinese
Control Console · English
Group-Control Router Dashboard · Devices & Pairing
Router Captive Portal · One-Tap Provisioning
Core Principles
Upgrading "multi-device action" from mechanical scripts to "natural-language driven + LLM parsing + carry-everywhere hardware pairing".
LLM Task Understanding
Turns "open Weibo, check trending and summarize for me" into a real step sequence (CLICK / BACK / SWIPE / INPUT); the same LLM observes the screen mid-run and revises upcoming steps as needed.
20-¥ ADB Group-Control Router
A self-built ESP32 router persists multi-device ADB pairing info — keys, ports, device ID → IP mappings — in local storage + TLS keystore. Switch Wi-Fi, switch office, switch city — no re-pairing; all pairings travel with the router.
Live Screen Stream + Chain of Thought
Live monitor lines up "device screen stream", "LLM Chain of Thought" and "step timeline (CLICK @30,79 / BACK @175,380 …)" in one view, with prev / next / follow-latest switching; failed steps are immediately tagged Failed.
Replayable History Sessions
Every task (Completed / Failed / In-Progress) is persisted and indexed by device / time / task name — a real replay source for ops, QA and customer-service retrospectives.
Industry Status & Pain Points
The more you scale multi-account ops, the more you get stuck on "lost pairings", "opaque scripts", and "locked-in models".
Manual Multi-Device Ops
Ops / QA / CS repeat the same click flow across N phones — wasteful, error-prone and uneconomic at scale.
Group-Control Routers Lock Networks
Commercial group-control boxes tie devices to a specific Wi-Fi / LAN; switching offices forces complete re-pairing — nothing is "take-it-with-you".
Closed-Source Model Lock-In
Most solutions bind to a single paid closed-source model; swapping to a cheaper / sharper model means rewriting the whole integration.
No Replayable Operation
Existing scripts only have a start / end; no screen stream, no thought chain, no prev / next "recording" — when something breaks, you can only guess.
App Compatibility Pain
Every new app launch demands re-running the same test script on every model and resolution; there is no engineering capability to "write once in plain English, run on N devices".
Group-Control Router · Pain Points & Self-Built Solutions
The 20-¥ ESP32 self-built ADB group-control router, mapped across six dimensions — pairing / addressing / provisioning / concurrency / security / cost.
| Pain Point | Solution | Key Module | |
|---|---|---|---|
| Lost pairing info: switching networks or devices invalidates all ADB pairings, forcing a re-flash of pairing keys from a PC. | NVS Keystore persistence: ADB pairing keys, device ID→IP map and TLS certs are written to ESP32 NVS — survive power loss; plug into a new network and go. | adb_pairing.c + nvs_flash | Pairing |
| Address chaos: multiple devices drift IPs and clash ports; the console cannot find or tells devices apart. | Unified local port map: the router runs an ADB-over-TCP reverse proxy, giving each device a fixed virtual port; the console only connects to one router address. | adb_proxy.c + tcp_forward | Addressing |
| One-shot provisioning: first-time setup needs a PC to edit code and flash — unusable for non-technical users. | Captive Portal + AP fallback: joining the router hotspot auto-opens a setup page; scan / pick Wi-Fi, enter password — no PC required. | captive_portal.c + wifi_manager.c | Provisioning |
| Unstable concurrent stream: N devices screenshot / ADB at once — bandwidth contention, drops, latency jitter. | Wi-Fi 60s supervisor + per-device health-check: auto-reconnect on drop, degraded mode on weak signal, heartbeat keep-alive. | wifi_manager.c + health_check | Concurrency |
| High ADB pairing security bar: pairing code in cleartext, no TLS — sniffable on the LAN. | Full AOSP pairing stack: SPAKE2 + Ed25519 + AES-GCM, per-device keystore, identical to official wireless debugging. | adb_pairing.c (TLS) | Security |
| High scale cost: commercial group-control boxes cost thousands per unit and lock you to one vendor. | 20-¥ BOM self-build: open-source ESP32 firmware, mass-producible and modifiable — locked to no commercial hardware. | main.c + custom firmware | Cost |
Core Capabilities
Upgrading multi-device collaboration from "script automation" to "natural-language driven + visible-in-real-time + replayable".
Natural-Language → Actions
Describe the goal in plain Chinese / English; the LLM auto-produces executable steps (CLICK / BACK / SWIPE / INPUT) and continuously reflects on screen captures for the next step.
20-¥ Carry-Everywhere Router
Self-built ESP32 firmware as the group-control gateway; ADB keys, IP mappings and TLS certs live in NVS — plug into a new network and you are back online.
Multi-Device Concurrent Run
Select N target devices when creating a task; each device runs an independent session with its own step timeline, all aggregated into one monitor view.
3-Track Live Visualization
Screen stream + LLM Chain of Thought + per-step timeline side by side; prev / next / follow-latest switch freely so every action is explainable.
OpenAI-Compatible Models
Default model gelab-zero-4b-latest; compatible with any OpenAI-style endpoint (OpenAI / Claude / DeepSeek / Qwen / Gemini …); detector / parser are decoupled so you pick per latency / cost / precision.
History + Failure Replay
Task results, per-step detail and Chain of Thought are persisted; success / failure / partial states are distinguishable, re-runnable, and exportable for ops retros.
Security & Audit
Each device / task gets an independent session; ADB TLS + per-device keystore; full operation audit logs for compliance and permission tracing.
Decoupled Parser / Model
The detector that turns language into steps is a standalone parser — unit-testable, swappable — so model upgrades no longer require a full regression.
From "One-Sentence Instruction" to "N Phones Acting in Concert"
Four console phases · natural-language driven + 3-track live visualization
Key Outcomes
Lowering the bar, unit cost and traceability of multi-device group control step by step.
Core Breakthroughs
Taking "multi-account group control" from mechanical scripts, locked hardware and unreplayable runs to "carry-everywhere / visible / explainable / model-portable".
20-¥ Carry-Everywhere Router
The self-built ESP32 ADB router bakes multi-device ADB keys, IP mappings and TLS certs into local storage — switching networks / offices / cities no longer requires re-pairing.
3-Track Real-Time Visualization
Screen stream + Chain of Thought + step timeline share one view; the LLM's "why click here" lives in the task stream — failed runs are replayable like a CCTV recording.
Model / Parser Decoupled
detector / parser / model are all independent — swap any OpenAI-compatible endpoint or parsing strategy; no more lock-in to a closed model or commercial group-control box.
Group Control + LLM, Engineered as One
Console + router + LLM parser live in the same repo as a tightly coupled system — "research prototype equals product prototype"; no second-pass integration needed.
Business Value
Convert technical capability into quantifiable, reusable business value.
Content Under Active Update
Product content has been published based on internal materials. The following areas are planned for further development:
Other Products & Solutions
Explore Xianma AI solutions in other domains