Canvas + built-in IDE
Objects, features and relations on an image. Screph Code provides an editor with a coding agent, a terminal and program execution controls.
Desktop application · Version in development
Screph: a canvas and IDE with an AI agent for computer vision
Object detection, text recognition and image-based automation. Canvas annotation, processing operations and code use shared project data.
Download Screph
Windows, Linux
Documentation
Visual builder
Sequences of computer vision operations built from ready-made blocks. Parameter controls and a preview of each step's output.
An agent whose results you can inspect
Creates annotations, features, relations and processing workflows using standard tools. Changes appear on the canvas and in panels; you can inspect and correct the result.
Custom panels and processing
Plugins add panels and commands. Python operations support custom processing. You can create an extension with the agent.
From a frame to a program
Annotation, processing and code are available in one environment. The agent can participate at each stage; operations and results can be inspected in the interface.
| Stage | Work in the application | Output |
|---|---|---|
| 1. Canvas | Frame selection, object annotation, features and relations, created manually or with the agent. | An image and a description of what to find, read or compare. |
| 2. Computer vision | Build a workflow from ready-made blocks without code. Set parameters, preview steps and test on other frames. | A processing workflow and its results: masks, contours, text or detected objects. |
| 3. Screph Code | A developer or coding agent adds program logic, edits files and tests execution. | Code to repeat processing, use its results and perform actions. |
Finding errors requires varied examples, including frames without the target object. A successful result on one image does not guarantee accuracy on others.
Tools already available in Screph
Frame capture, annotation, processing and the code editor share a project. Each stage has its own panel.
| Task | How Screph handles it |
|---|---|
| Capture frames for testing | Image and video files; screen, window or camera capture. A timeline for selecting a frame. |
| Describe an object and its search area | Annotate directly on the canvas: geometry, features, descriptions and relations. The agent can create and edit them. |
| Build a processing workflow without writing code | Ready-made operation blocks, parameter settings and a preview of each step result. The agent helps build and modify the workflow. |
| Connect recognition to program logic | Screph Code uses project materials. The editor, chat, terminal and run output are available side by side. |
| Add a tool for your task | Custom panels and commands through plugins; custom algorithms through Python operations. |
Finding a part: annotation, processing, code
Demonstrations using one image: describing an object, comparing computer vision operations and writing a search program.
Object, search area and visual reference
We define the search area, select the part template and mark a separate visual reference on the image. We refine the box, check undo after moving it and link the template to the reference. Descriptions preserve the purpose of the marked areas.
Loading...
Comparing two processing workflows
First, we extract edges with a single Canny operation: both the part and speckle noise appear in the result. Then we run a Median Blur → Canny chain and compare the masks. Smoothing first removes unwanted edges while preserving the part's outline.
Loading...
Code and test examples
In Screph Code, we open a prepared project with annotations, instructions and processing materials. Then we inspect the agent's task and the test results for three cases: the part is in place, moved or absent. The agent response and run results in the web demonstration are prepared in advance.
Loading...
Capabilities and workflows
Open map separately
How Screph tools connect
The map shows how capture, Canvas annotation, computer vision and program development connect to the agent, model training and other tasks.
Loading graph…
Connections between modules
Each icon represents an action. The highlighted line connects capture, Canvas, computer vision and Code. Related workspaces sit nearby. Hover over an icon to read its description.
Hover for a description · Drag to move · Scroll to zoom · Space to pause
Custom tools inside Screph
Plugins add custom panels and commands to application workspaces. You can create them with the built-in Agent and load them from a personal folder. Python operations are available for custom image processing.
From a task description to an extension
Extensions → Create with Agent prepares a folder with a development kit and a task draft. After the request is submitted, the Agent creates the extension; connecting it is a separate action.
In this example, the agent creates a panel that filters parts by the longer side of their bounding box in pixels. A two-handle slider sets the size range: six parts are narrowed down to three, then two. Plugin settings open in a separate window.
Interface demo
Screph workspace in the browser
The demonstrations show part annotation on Canvas, CV setup, the agent working with a frame and project handoff to Screph Code. This browser-based learning interface explains the steps using prepared data, without running your files, models or devices.
Tools and project data
Panels for capturing frames, describing objects, configuring processing and developing a program. Ready-made operations are configured visually; custom logic is added in code.
- Capture and timeline
- Images, video and live sources
Import files and frame sequences, or capture a screen, window or camera. Select a frame for annotation and processing checks.
- Canvas
- Objects, features and relations
Rectangles, ellipses and polygons; search areas, templates and color features. Descriptions and relations preserve the meaning of annotations.
- Computer vision
- Operation blocks and preview
Color processing, filters, contours, OCR, and object detection and segmentation models. Ready-made blocks connect into a workflow without writing code; previews show intermediate and final results.
- Screph Code
- Built-in IDE with a coding agent
Files, editor, chat and terminal. Project materials provide context for writing code; you can run the program and inspect its output.
The agent's work is visible in the interface
The agent uses standard Screph tools to create the same objects as the user: annotations, features, relations and processing workflows. You can open, inspect and edit them in the application.
Annotation and relations
Propose bounding boxes from an image; create and edit elements, search areas and features. Move, align, rename and link objects.
Processing workflows
The agent builds chains and graphs from operation blocks. The workflow, parameters and step results are available in the computer vision panel. For custom processing, it can prepare Python code for a separate step.
Project and program
Save a project, work with open documents and hand off a task to Screph Code. In the IDE, a coding agent writes and edits the program using the selected materials.
A configured model connection and access to the required tools are needed. Access settings govern image transmission and action confirmation; frame analysis requires a model that supports images.
One frame or a set of examples
Object, text or state
Find an icon, read a number, isolate a part by color or compare frames. You can repeat the selected processing on other examples and examine errors using images and step results.
A model trained on your examples
Annotate images, classes and masks, prepare a dataset and start training. Classification, detection and segmentation tasks are supported; a compatible model and training environment are required.
Screph is in development. The status page describes the readiness of individual tools and current limitations.
Application features
Interface and tools
Canvas, capture, computer vision, agent, Code and settings. Screenshots of the version in development can be opened at full size.
Toolbar
Object type and box shape are selected separately: element, feature or area; rectangle, ellipse, polygon or lasso.
Select multiple boxes
Selection tools collect several boxes into a temporary selection. Ctrl-click adds an object to or removes it from the selection.
Information panel
The “Graph” and “Tree” tabs show objects and their relationships. The graph helps you understand the annotation structure.
CV setup
The “Build workflow” section selects an operation or a chain of steps. For example, Median Blur removes speckle noise before Canny detects edges.
Recalculate and save
“Run and save” computes the result using the current parameters. The saved processing result remains separate from annotations.
Image histogram
The inspector shows pixel-value distributions by channel. The histogram helps assess an image before configuring CV.
Capture panel
The “Source” section selects the frame source, recording mode and capture area. These settings determine which image enters the workflow.
Object features
The table shows the image crop, mask and feature preview with their statuses. You can open or save the files again from here.
Agent conversation
Use the request field and “Attach Canvas frame” button to ask the agent. A configured model connection is required for a response.
Dataset preparation
The “Next step” section offers to create or select a dataset and open advanced settings. Prepare examples before training.
Canvas minimap
The minimap shows the whole image and the visible area's position. It helps you navigate a zoomed Canvas.
Playback controls
Navigation, play and stop buttons control frame viewing. Playback speed and looping are configured nearby.
Timeline scale
“−” and “+” change the scale. “Fit to window” and “Fit to frame” help choose a suitable timeline view.
Graph filters
The “Ownership” and “Relationships” checkboxes control the two relationship types displayed in the object graph.
Dataset actions
Nearby commands create a dataset, add data and open an example. Available actions depend on the selected dataset.
Project explorer
The folder and file tree helps locate project materials. The filter field narrows the list when you search for a file.
Run code
“Run”, “Stop” and “Repeat” sit together. Use them to control execution of the prepared code.
Output and terminal
The tabs below the editor switch between output, terminal and plan. Work panels sit beside the open file.
FAQ
Before your first task
Do I need to know programming and computer vision beforehand?
Annotation and ready-made operation settings are available through the interface. The agent helps choose processing operations and write code. Custom logic and integrations may require programming; results are tested on examples from the task.
What do I need to connect to use the agent?
Choose a language model connection in settings. Image analysis requires a model that supports images and permission to send the frame. Some computer vision methods require their own models and computing components; the documentation lists these requirements.
How does Screph work with code?
Screph Code combines an editor, project files, a coding agent chat and a terminal. Programs can be started and stopped here. Selected images, annotations and computer vision results provide context for the coding agent.
Is internet access required?
Annotation, project saving and traditional processing operations work locally. Cloud agents, speech services, model downloads and some integrations require internet access.
Are projects uploaded to the cloud?
Projects and images are stored locally. A cloud model receives request materials within the configured access scope. Separate permissions govern image and file transmission.
Do I need to install additional components?
OCR requires an engine and language packs. Model-based operations require weights and a compatible computing environment. The components needed depend on the operation.
Download and documentation
Screph in development
The homepage describes the tools in the development version. The download page lists available builds, installation steps and requirements. The status page covers limitations of individual features.