jev-mobile

# jev-mobile

[](https://www.python.org/)
[](LICENSE)
[](#project-status)

An autonomous, durable Android sub-agent with [TypeSafe Jev](https://typesafe.ai/) as its System-One decision provider.

`jev-mobile` accepts a high-level task, persists it in SQLite, and lets one durable worker execute **observe → normalize → decide → mutate → verify** against a USB-connected Android device. Jev receives concise semantics and technically valid actions; it never generates coordinates, MCP calls, or arbitrary code.

> **Safety boundary.** Durable Core V1 has been hardware-validated for normal notes/checklists, text input, crash recovery, persistence verification and foreign-editor protection. It deliberately stops rather than clearing unrelated application state. Payments, account changes and other sensitive operations remain out of scope without explicit policy/approval work.

## Why this exists

Traditional mobile-agent stacks repeatedly involve a large model in every UI step:

```text
LLM → mobile tool → LLM → mobile tool → LLM
```

The high-frequency loop remains local and structured:

```text
High-level goal
       │
       ▼
┌────────────────────────────────────────────────────┐
│ jev-mobile                                         │
│ durable worker: observe → normalize → requirements │
│ → valid actions → Jev → journal → verify           │
└────────────────────────────────────────────────────┘
       │
       ├── Portal ADB / USB Android runtime
       └── Escalation for uncertainty or risk
```

For example, Jev sees only this decision problem:

```text
Task: Create a note titled Shopping

Valid actions:
  A1  Create entity
  A2  Set title in title field
  A3  Verify persisted result

Jev → A1
```

The worker maps `A1` to the backend through the DeviceAdapter. Invalid actio