AI, Screph Code and automation
These parts of the product use project data differently and run separately. Saving the project does not itself issue a LLM request, change the code, or perform any input actions.
Automation Runtime Implemented
The Automation Runtime library provides a prepared program with visual and text matching, waits, and a unified input API. Automation Manager and the Run and check window launch the program separately. The input method is selected in settings; the script runs with the permissions of the application that launches it.
Preparation and safe launch →Agent Experimental
Offers explanations and supported changes in the context of the current task. The feature is off by default. Step-by-step mode presents changes for confirmation; delegated mode continues the permitted task until it reaches a result, is stopped, or needs user input.
Context, permissions and history →Screph Code Implemented
Screph Code is the built-in agentic IDE. It receives selected project data and a code working directory. Builder can edit files and run permitted commands in that directory. Created, modified, and deleted files can be reviewed in the editor and changes panel; a backup or version control provides a way to revert.
You can run a file manually with a separate command. Builder can also execute code during a task through permitted terminal commands.
Working with Screph Code →Screph Code working directory and session
Agent parameters offer By connection, Codex, and OpenHands. The agent engine, model, and access settings are checked before a request; a connection error does not switch engines automatically. Files, project objects, conversation history, and terminals belong to the selected working directory.
Unsaved text stays in the editor until it is saved or explicitly discarded. Closing can be cancelled. External file changes require review. Stop interrupts the agent's work but does not undo files already written; background commands may take time to finish stopping.
Codex Desktop opens as a separate application. Transferring materials creates a task draft; submitting and stopping it are handled in Codex Desktop. Its account and history are separate from the Screph chat.
Optional ROS 2 integration
The Settings and tools → Tools → ROS 2… panel offers three levels. Opening the panel does not start ROS. The coding level prepares a new working directory with instructions and selected saved project materials; this preparation does not require ROS.
The exchange level supports frame files without ROS and, after connecting, can list topics, take a snapshot, and explicitly publish an image. A snapshot is sent to Capture with a separate command. The continuous level displays the latest stream frame; processing with a saved scheme is enabled with a separate checkbox and requires a project and a selected object with a scheme.
Topics require a local Python interpreter from a prepared ROS environment with rclpy and the message types. ROS and its system libraries must be installed separately before Screph starts. After settings are applied, connecting and starting the stream are explicit actions. Stopping disconnects it; continuing requires a new connection.
Code export creates a new directory and does not overwrite an existing one. The target ROS version, operating system, and final package checks are specified in the coding task. The integration has not yet been validated on a real DDS network; a topic list does not guarantee frame delivery. It does not provide robot navigation or motion commands.
Local and cloud models
Local Ollama and LM Studio profiles process permitted context in a local service. Cloud profiles send context to the selected provider. Profiles are assigned separately to Screph Code, the agent and other functions.
Connections and data transfer →Pre-flight checklist
- The project and working files have been saved.
- Checked what data and images will be transferred.
- An available model and the required dependencies have been selected.
- A result check and a way to undo changes are ready for the action.