Chat is where Agent receives context, proposes plans, runs work, and returns deliverables.
Inputs Agent can use
Agent can use text prompts, transcribed voice text, uploads, project files, previous conversation context, and explicit user constraints.
Text prompts
Audio-to-text input
Uploaded files
Project files
Previous artifacts
Plan approval
When the task is heavy, Agent can show a concise plan before execution. The plan should name the goal, expected deliverables, main steps, and task type.
After approval, Agent should follow the plan but still adapt when it discovers better evidence or a required tool.
Research -> collect sources -> compare -> create DOCX.Frontend -> inspect app -> implement -> run checks -> preview -> fix UI.Outputs and artifacts
Final answers should explain what changed and link generated artifacts. Files remain attached to the conversation and project so they can be reused later.
Result message
Generated files
Preview links
Logs or test summary
Next useful action