The EnterpriseClaw: Run Agent action runs an agent loop.
Procedure
No Validate Workstation step is required. Minimum bot agent version:
22.180
-
Select the Model Connection provider for the agent
loop.
The following models are recommended.
Provider Model Amazon Bedrock - Claude Opus 4.5
- Claude Sonnet 4.5
Azure OpenAI - GPT-4.1
- GPT-5
- GPT-5 mini
Google Vertex - Claude Haiku 4.5
- Claude Opus 4.5
- Claude Sonnet 4.5
OpenAI - GPT-4o
- GPT-4o mini
- GPT-4.1
- GPT-5
- GPT-5 mini
-
Specify the Goal / Task. Two input modes are
available:
- Specify a path to an external Goal/Task instruction File.
- Enter a natural language prompt as Text.
- Enter the number of Max turns the agent should make before time out.
- Optionally, select a Workspace directory path for the agent to manage tools and capabilities.
-
Select the tools for the agent to use during run time.
Option Description Tool Description Spreadsheet Read, write CSV and MS Excel files PDF Reader Extract text from regular and scanned PDFs Web Content Read webpages, invoke APIs, download files Computer Use Control the screen via click, scroll, type Browser Use Open, read, and interact with a browser Custom Tools Upload a JSON file with custom tool definitions -
Configure Capabilities of the agent runtime behaviors.
All three capabilities default to enabled:
Option Description Capability Description Memory (required) Persistent memory across runs. This capability is always on and cannot be disabled by the user. Validate Goal/Task Completion Agent verifies completeness of output before finishing the loop. Verbose Logging Logs reasoning steps, tool calls, token usage, and progress. Useful for debugging and demos. - Optionally, provide credentials the agent needs for tool execution (e.g. API keys, login secrets), using the Add Credential button.
- Store the output in a variable with one or multiple variables, using the Assign result to field.