Screph Documentation
Practical documentation for the complete program: from a source image or video, markup and voice descriptions to a structured task for AI, an agentic IDE and execution. Computer-vision methods are optional; statuses and limitations are called out wherever they affect the process.
Quick path
Install and launch
Requirements, first launch and basic configuration.
02Open source data
An image, video, screen region, window, camera, stream or media set.
03Create and describe markup
Elements, annotations, voice descriptions, relations and groups.
04Apply computer-vision methods when needed
Methods, linear/graph pipelines, derived data, candidates and diagnostics.
05Save the task
A main project, images, attached artifacts and a for_ai_agent context.
Continue the work
The agentic IDE, supported external tools, AI Assistant and Automation Runtime.
Configure the environment
System requirements
Windows x64, CPU and RAM boundaries, optional CUDA, disk gates, devices, network and permissions.
Dependencies and updates
Runtime sources, probes, repair, signed ML packs and component-aware updates.
Windows and permissions
Installation, UAC, Virtual HID, Arduino firmware, hotkeys, system settings and uninstall.
LLM and speech recognition
Connections, model roles, local operation, voice providers and privacy.
Account and online services
Local boundaries, device login, server speech, support, diagnostics and the current payment status.
Integrations and extension
JSON and API boundaries, CLI, semantic catalogs and the boundary between internal registries and external plugins.
CV models and weights
YOLO, SAM and OmniParser runtimes, storage, downloads, active models and SHA status.
Local data and secrets
ScrephData, settings, Credential Manager, history, caches, retention and complete cleanup.
Automation Runtime
Project loading, screen/OCR matching, input backends and Automation Manager.
AI Assistant
Activation, workspace, competencies, context, action cards and confirmation boundaries.
Settings map
Where to configure modes, models, dependencies, input, IDE, voice and diagnostics.
Diagnostics and recovery
Crash sessions, bundle contents, local export, startup recovery and manual submission.
Understand the data model first
If you are new to Screph, read about Source, ROI, the main project, candidate/evidence and the distinction between Relation Graph, Method Graph and Action Trace.
Open core concepts →Documentation follows the code
Screph is under active development. Before a critical workflow, check the current status and dependencies of the selected component.