Create an Agent
Creating an agent is a conversation. You describe what you want in plain language, the planner asks clarifying questions if needed, and then presents a plan for your approval.
You do not need to manually select skills, wire steps together, or write code. The creation system designs the pipeline based on your description.
Opening Tool Forge
From the Tools page in ThinkForge, click Create Tool. This opens Tool Forge — a dedicated creation workspace for defining the tool through conversation.
Step 1: Describe what you want
Type a natural-language description of what the agent should do. Be specific about:
- What kind of content it will process
- What transformation or analysis you want
- Where the output should go (saved file, displayed on screen, sent to a service)
For example: "Make an agent that takes meeting notes and produces a summary with action items, then saves it as Markdown."
Step 2: Clarifying questions (if needed)
If your description is ambiguous or could go several directions, the planner asks clarifying questions — typically 1 to 3 at a time. These are about the agent's purpose, not about file paths or input format (you provide input separately when running the agent).
Answer in the same chat. The planner may ask follow-up questions before proceeding.
Step 3: Plan review
Once the planner has enough information, it produces a refined plan and shows you a summary:
- Agent name and description
- Number of steps in the pipeline
- Capabilities the agent will use
- Memory scope (session, project, or global)
Step 4: Approve or reject
An approval panel appears with the full plan details. You have two choices:
- Approve — the agent is validated, saved to your library, and ready to deploy
- Reject — go back to the chat and refine your description
On approval, ThinkForge validates that all proposed capabilities exist and the plan structure is sound, then saves the agent.
After creation
Your new agent appears in the Tools library. From there you can:
- Deploy it (add to Global or Current Project scope)
- Configure input sources, reference files, and output folder
- Adjust settings (execution mode, output format, instructions)
- Run it immediately or set up triggers
What you control vs. what the system handles
- You specify — what the agent should do, answers to planner questions, approval
- The system handles — which skills to use, step ordering, data flow between steps, validation
You do not manually pick skills from a list. The planner selects from an approved set of capabilities based on your description. You see the chosen capabilities in the approval summary.
Editing after creation
After an agent is created, you can adjust its behavior through Settings — instructions, output file format, execution mode (guided vs. autonomous), and API targets. You do not rearrange the internal step pipeline directly; instead, refine behavior through the settings interface or create a new agent if the workflow needs fundamental changes.
Related
- Agents Overview — What agents are and how they work
- Run Agents — How to execute your created agent
- Permissions — What capabilities agents can have