Operator

Local Android assistant that runs JSON-defined workflows through accessibility actions and stores outputs as artifacts.

What We Did

• Implemented a workflow engine (WorkflowRunner) with JSON parsing (WorkflowJsonParser).
• Added artifact storage with deterministic naming (ArtifactStore, ArtifactNamer).
• Added workflow queue storage using SharedPreferences (WorkflowQueue).
• Added accessibility automation (OperatorService) and a sample workflow (chatgpt_haiku.json).
• Added unit tests for JSON parsing edge cases in app/src/test/java/com/operator/app/workflow/WorkflowJsonParserTest.kt.
• Added an in-app note and build doc note about local quantized model files being user-provided.

Local Quantized Model Note

Local quantized model weights are not bundled in this repo. Each user must download and install their own model files for their chosen local LLM runtime. Keep model paths in local configuration and do not commit model files.

Memory Storage

• Artifacts: stored in app-internal storage under context.filesDir, using paths like Operator/Projects/<project>/<workflow>/....
• Workflow queue: stored in SharedPreferences under operator_workflow_queue.

Build

./gradlew assembleDebug

Tests

./gradlew testDebugUnitTest

Project Layout

• app/src/main/java/com/operator/app/accessibility/: accessibility automation
• app/src/main/java/com/operator/app/workflow/: workflow models + runner + JSON parser
• app/src/main/java/com/operator/app/storage/: artifact storage + queue
• app/src/main/assets/workflows/: sample workflows