1. Create a project
- Open Agent Studio and choose New project.
- Name the project and optionally pick a starter template from the marketplace.
- You land on the visual canvas with a blank (or templated) agent graph.
2. Compose agents on the canvas
Drag agents, tools, and decision nodes onto the canvas. Connect outputs to inputs to define control flow. Use branching, parallel paths, and human-in-the-loop nodes when a step needs approval.
3. Connect MCP tools
- Open Tools → MCP.
- Add an MCP server URL or local command for your integration.
- Scope permissions per tool, then attach tools to the agents that need them.
4. Configure prompts and memory
- Use the Prompt library to version system and task prompts; roll back if a change regresses quality.
- Enable short-term and long-term memory stores with per-agent or per-user scoping.
5. Add retrieval (RAG)
Create a knowledge source, upload or connect documents, and wire a retrieval node into your workflow. Hybrid search and citation tracking help keep answers grounded.
6. Run and inspect
- Click Run to execute the graph with sample input.
- Inspect run history, tool calls, and intermediate state for debugging.
- Replay a prior run when diagnosing intermittent failures.
7. Export and share
Export the workflow definition for version control, or publish a template to your team's marketplace. See features for the full capability list.
Need help?
If installation is incomplete, return to the install guide. For product demos or enterprise support, contact VynelixAI.
