Core concepts

Screph concepts: sources, annotation, processing, projects, training, code, and connections. Short interface labels are clarified by their section context.

Sources and capture

Data source: An image, video, screen, window, camera, stream, or other source of visual data. In capture settings, this is Capture Source.
Frame: An individual image associated with a source and its position in a sequence.
Frame sequence: Source frames ordered in time. A frame range selects part of the sequence.
Capture area: The part of a screen or other source from which Capture obtains pixels.
Capture recording: Saving frames and events supported by the mode. Timed recording and input-event recording differ in how frames are captured; memory and disk storage are configured separately.
Action trace: A sequence of user actions and associated frames. Visual targets proposed during its analysis are checked separately from the original recording.
Source image: The source image on which the work is based. Processing may use a different representation of this image.

Annotation and relations

Annotated element: A saved task object, such as a button, field, or part. Called Element in the tools.
Element feature: A property or geometry belonging to an element, such as an area used to locate it or a visual sample. Called Feature in the element panel.
Annotation: A description of an annotated object, its purpose, or use. Voice annotation converts speech into a text description of the selected object.
Element relationship: A domain relationship between annotated objects. In a processing graph, data transfer is called a connection.
Container area: A structural annotation area with a defined set of objects. Being inside its boundary and belonging to its set are tracked separately.
Selected objects: The current selection for an action. It may contain several objects and change without changing container-area membership.
Current object: The object whose properties or actions are currently shown. When several objects are selected, the current object provides context for a specific action.
Annotation proposal: Proposed objects or changes that can be reviewed before acceptance. Inspecting a proposal does not change the main annotations.
Interaction set: Selected Canvas interaction rules: click behavior, dragging, modifier keys, and initial tool.

Practice example. Changes apply only here and reset when you leave the example.

A feature area is selected separately from an element. The example shows its geometry and properties.

Computer vision and results

Processing operation: An available action with defined inputs, outputs, and parameters, such as edge detection or text recognition. An algorithm or model determines how it runs.
Processing step: A configured instance of an operation in a workflow. The same operation can appear in several steps with different parameters.
Processing workflow: A configuration of steps, connections and selected outputs. A workflow can contain one operation, a chain or a graph.
Processing chain: A linear workflow: steps run in order and pass data to subsequent steps.
Processing graph: A graph with explicit input and output connections. It can describe branching and merging data.
Graph node: An object in a processing graph. An operation node represents a step; other nodes can define a source or result output.
Input and output: Named input and output points for data of a defined type. For example, an operation may take an image and produce a mask.
Connection: Passing data from a specific output to a specific input in a processing graph.
Step parameters: Values that control a specific step, such as a threshold or filter size.
Input data: The images, text or other values passed to a workflow or step during a run.
Processing area: The geometry used as processing input: the whole image or a selected part. It is defined separately from the capture area and the container area.
Processing run: One workflow execution with fixed inputs and settings. Repeating it creates a new run.
Processing result: Result data: an image, mask, contour, detection, text, or metric. Moving detected objects into the main annotations requires separate confirmation.
Step result: Intermediate data from one step. Viewing it does not make it the workflow’s final output.
Preview: Displaying data for evaluation before the next action. Save the result separately unless the command explicitly combines running and saving.
Overlay: A mask, contour, or other data shown over an image for inspection.
Saved result: The fixed data from a specific run. Changing inputs or a graph may make it outdated for the current work.
Result materials: Images, files, and other resources associated with the resulting output.
Operation parameter set: Saved parameter values for one operation, for reuse.
Saved operation: A named custom operation with code or instructions, inputs, and outputs.
Workflow template: Saved steps and connections for reuse. Loading a template prepares a workflow.
Workflow draft: The current editable processing configuration before the selected save or run action.
Workflow variant: An alternative configuration for comparison, with different parameters, steps, or connections.
Experiment: A saved comparison of processing variants on reference inputs, together with their results.
Base variant: The selected point for comparing variants. Execution comparisons use a baseline run.
Workflow fragment: A reusable part of a processing workflow.
Python code: A custom operation with saved Python code, connected inputs, and declared outputs. Generating code and running it are separate actions.
AI request: An operation that sends permitted inputs and instructions to the selected model and checks its response against expected outputs.

Practice example. Changes apply only here and reset when you leave the example.

The chain links image filtering and edge detection. Parameters belong to the selected step.

Project and interface

Main project: Editable Screph task data: sources, annotations, settings, and related materials. Transferring a saved project usually requires its directory.
Project folder: The location of the project and related files on disk.
Prepared project materials: Documents, indexes, and resources generated from project data for coding tools and other consumers.
Project snapshot: A fixed project state for a specific purpose, such as passing it to coding tools.
Project export: A prepared representation of selected data for transfer or an external tool. Its contents depend on the selected format.
Workspace: A set of tools for specific work, such as annotation, training, or coding.
Canvas: A visual workspace for images, annotated objects, and views.
View window: A place where content is displayed on Canvas. One workspace can contain several view windows.
View: View contents: a source image, processed image, mask, or comparison.
Display method: A display transform, such as grayscale, inversion, or pseudocolor. The processing input is selected separately.
Panel layout: The geometry and visibility of interface parts. Saving the layout applies to the interface.
Program mode: A task mode, such as General or GUI Automation. It sets the context for tools and tasks.
Simplified Mode: An interaction method that reduces intermediate actions and handoffs between existing tools.

Datasets and Training

Dataset: A managed set of data and annotations for the selected training task.
Dataset sample: One data record with its own provenance. It contains the sample image.
Dataset annotation: Accepted sample annotations for the selected task. Importing or saving alone does not establish human review.
Annotation class: A category of dataset objects, such as a part type.
Dataset draft: Editable images, classes, and annotations. Make corrections in the draft.
Dataset version: A fixed snapshot of data and annotations for training or evaluation. Draft changes belong to a new version.
Training configuration: A consistent training description: data, model, and execution parameters.
Training parameter set: A partial configuration preset, such as Fast, Balanced, or Thorough. Data and model selection remain separate.
Experiment: Organizing training configurations and runs for comparison.
Start training: One training run with fixed inputs and configuration, history, and results.

Models and their data

Language model: A model selected to work with text and, when supported, images or other data. It produces responses within the supplied input.
Model architecture: The family and structure of a trainable model.
Model weights: Parameters of a trained model. Pretrained weights serve as initial data for supported training or processing.
Model candidate: A model produced by a specific training run for evaluation.
Released model: A registered model with fixed weights and provenance information. Its validation results are tracked separately.
Primary model: A released model selected for a specific dataset task. Use in the final program is enabled separately.
Model-use profile: A description of a prepared model and its role in a specific task or final program.
Calibration profile: Calculated and validated camera parameters. Measurements in real-world units also need a suitable spatial reference.

Code and IDE

Screph Code: A built-in agent IDE for working with code files, project context, agent conversations, and program checks.
Code workspace folder: The directory of files used by Screph Code or an external editor. Access to this folder defines the boundary for coding work.
Script: A specific program executable.
Task for a coding agent: A task with a request and project context sent to an executor. Delivery, acceptance, and completion are separate stages.
Pass to Code: Preparing a saved project revision and task for a supported coding tool. Agent execution is controlled by a separate setting or explicit submission.

Practice example. Changes apply only here and reset when you leave the example.

The working folder contains prepared materials and program files. The learning example shows how to select and inspect them.

Agent and conversation

Agent: An interactive executor for requests in Screph context. It can make a plan and perform permitted actions with available tools.
Agent executor: The component that executes agent tasks and provides its tools. The model and execution environment are configured separately.
Conversation: Saved conversation history with an agent in the selected context.
User request: A request containing text and attachments. Fulfilling it may require several model calls by the agent.
Model prompt: Prepared model input: instructions, context, and data. It may include a user request and other messages.
Plan step: An individual item in an agent plan.
Agent action: Reading, running, or changing something for a defined purpose. A change may first appear as a proposal for review.
Agent skill: Instructions and related material for performing a specific task. A skill describes the steps; available tools and permissions are set separately.

Execution and runtime environments

Script: A description of behavior or a sequence of actions, such as an automation or validation script.
Run configuration: Prepared inputs and settings for a specific execution.
Run: One execution of a task: running a computer vision workflow, training a model or executing a script.
Runtime environment: An installed environment with dependencies for the selected execution type.
Environment component: A managed installation unit in Components and Environments. Whether an action is ready also depends on its other requirements.
Dependency package: A library or package included in an environment component.
Run and verify: The Runtime Workbench workspace for starting, observing execution, and inspecting results. Available actions depend on the selected executor.
Output: Data for a specific task: the result of processing, training, recognition, or running a program.
Supporting evidence: Reference frames, metrics, logs, and other observations used to evaluate a result.

Settings

Settings: A user-defined value with a defined scope and time of application.
Configuration: A consistent set of values for a specific object, such as a run, connection, or environment.
Settings profile: A named set of settings with a stated purpose. A connection profile and a model-use profile describe different objects.

Connections and speech recognition

Model provider: A service or implementation that provides access to models.
Connection profile: Saved access and model-selection settings. The list of configured links uses the short label Connection.
Model purpose: Choose a model for a specific task: processing, an agent, or code generation. A separate conversation may use a model other than the default.
Account: Identity and authorization with a specific service. Screph accounts and external model-provider accounts are handled separately.
Speech Recognition: Converting audio to text with the selected recognizer.
Recognized text: A text result of speech recognition that can be used for an annotation or request.
Recognition session: One session of audio recording and speech recognition, from start to finish.
Recognition log: A sequence of recognition events and results available for inspection or export.