AI Agent tools
AI Agent Builder
Plan what an AI agent should do, what information it can use, what it may act on, when a person should approve something, and how to test it—before choosing where to build it.
Quick starts
Choose one of 12 ready-to-customize examples, or browse the full library for a more specific task.
Build your blueprint in 3 steps
Describe your agent
Start with the job you want the agent to do, then add the information, tools, and work steps it may need.
Name & purpose
Give the agent a clear name, explain the job it should do, and define what success looks like.
Advanced: blueprint ID
An internal reference used in the downloaded blueprint. It is generated from the agent name until you edit it.
When should this agent start?
Choose what starts the agent and describe the situation that should activate it.
What information will the agent receive?
Add the information the agent needs to do its job, such as a request, document, spreadsheet, form submission, or record.
What information can the agent use?
Add reference material the eventual agent may consult, such as approved documents, websites, databases, search, or human-provided guidance.
How should the agent behave?
Describe the role it should take, the instructions it must follow, what it is responsible for, and when it should hand work to a person.
What tools can this agent use?
Add tools the eventual agent may need and decide whether it can only view information, make changes, or take an outside action.
How should the agent do the work?
Add the work steps the agent should follow and explain what should happen when a step fails. Technical step links are available under Advanced settings.
Advanced: first workflow step
Internal reference for the first workflow step in the downloaded blueprint.
Next, decide what the agent may do on its own and when a person should step in.
What decisions can the agent make?
Define situations the agent may encounter and what it should do in each one.
When should a person approve an action?
Add checkpoints for actions that should not happen automatically.
What must the agent never do?
Set limits that keep the agent inside its intended role, permissions, privacy rules, and safety boundaries.
What may the agent remember?
Decide whether the eventual agent needs memory and what information it may or may not retain.
What should the agent produce?
Define the reports, messages, records, recommendations, or other results the agent should create.
What should happen if something goes wrong?
Decide when the agent may try again, when it should stop safely, and when it should ask a person for help.
How will you test the agent?
Define realistic situations to test before trusting the agent with actual work.
Where could you use this blueprint?
Mycelgrid creates the plan for the agent. It does not connect accounts, deploy the agent, or run it for you.
This tool creates a planning blueprint only. Execution and external connections remain off.
When the plan and boundaries look right, build the blueprint to review and download it.
Review the plan in the preview. You can go back and change anything before downloading it. Advanced users can also edit the blueprint JSON directly.
Choose JSON or YAML for implementation handoff, or Markdown or Text for an easier-to-read copy.
Advanced: edit blueprint JSON
Uses schema mycelgrid.ai_execution_blueprint.v0.1. If you change the form after building, build again before downloading.
Need to change the plan? Go back, update your answers, then build again.