Overview
Davinci uses AI agents to automate complex engineering tasks within your system models. Agents can generate requirements, design system architectures, create documentation, perform analyses, and more—all through natural language instructions.
Each agent operates within its own workspace, maintaining separate chat histories and contexts. This allows you to work on multiple aspects of your model simultaneously with different agents, each focused on specific tasks or areas.
Working with Agents
For detailed information about the agent chat interface and input options, see the Agent Chat documentation.
Starting and Stopping Agents
- Play button: Starts the agent or resumes execution
- Stop button: Pauses agent execution
Stop an agent to add more context, provide clarifications, or redirect if it has gone off track. The agent will resume from where it stopped when you press play again.
Providing Context
The more specific you are with your instructions, the better results you’ll get. Provide enough details so the agent has all the context needed to perform its actions effectively.
Ways to provide context:
- Text instructions: Describe what you want in detail
- Images: Paste images directly into chat (png/jpg/webp formats) for visual reference
- Model objects: Reference specific objects using
@ followed by the object name
- Reference files: The agent can access uploaded reference documents and data
Web Search
Click the globe icon to enable web search for the agent. When enabled (colored blue), the agent can:
- Search the internet for information
- Import data, images, PDFs, and documents
- Add imported content as Reference objects in the Library
This is useful for pulling in specifications, standards, research papers, or other public information directly into your model.
Upload Reference Files
The Agent can interact with Reference Objects to get more information about your Model. Click the Upload a File button to add new Reference Objects.
Read more about uploading files to Davinci
Agent Personas
Agents can be assigned personas—predefined instruction sets that modify how the agent works. The currently selected persona is displayed in the dropdown menu.
Personas provide consistent instructions to the agent with each message, allowing you to customize agent behavior for specific workflows without repeating instructions. For example, you might have personas for:
- Detailed technical design work
- High-level architecture planning
- Documentation generation
- Requirements analysis
Personas can be created and managed in Project Settings.
Managing Agent Actions
Undoing Individual Actions
Each action performed by the agent can be undone independently. Click the Undo button next to any tool call in the chat history to revert only the changes made by that specific action.
This allows you to selectively keep the changes you want while reverting specific mistakes without losing all progress.
Restoring Checkpoints
Davinci automatically creates a checkpoint every time you send a message to the agent. This allows you to easily roll back to any previous state.
To restore a checkpoint:
- Scroll to the message where you want to restore
- Click the Restore Checkpoint button
- All agent changes made after that message will be undone
- Your message and any pasted images will be restored to the chat input for easy resubmission
Clearing Chat History
Click the trash can icon to clear the agent’s entire chat history. This starts the agent fresh with no prior context, which can be useful if:
- The agent seems stuck or incorrectly focused
- You want to start a completely new task
- The chat history has become too long
Clearing chat history cannot be undone. The agent will lose all context from previous conversations in that workspace.
Best Practices
Be Specific
Provide clear, detailed instructions rather than vague requests. Instead of “design a power system,” try “design a power system for a 3U CubeSat with solar panels generating 12W and a battery capacity of 40Wh.”
Use Model References
Reference existing objects with @ to help the agent understand context and maintain consistency with your existing model.
Provide Examples
Paste images of diagrams, sketches, or reference materials to give the agent visual context for what you’re trying to achieve.
Iterate Incrementally
Break complex tasks into smaller steps. Review each step before proceeding to the next, using checkpoints and Version Control to save progress along the way.
Review Agent Work
Always review what the agent creates. Use the Properties view, Tree view, and other visualizations to verify that the agent’s output matches your intent.
