Processors and code execution

Processors

Open Blueprint → Processors. Search for a processor or filter by purpose, then turn on the phase where it should run.

Input checks or changes incoming messages. Output checks or changes replies. Some processors also handle replies as they stream. Scoring evaluates the result.

Under Execution order, choose Configure processor to edit its settings. Use Move up and Move down to change the order within a phase. Use Remove processor to turn it off for that phase.

Start with one processor. Each adds work, and model-backed processors add model calls. Test the saved draft with a message that should pass and another that should trigger the configured behavior.

For example, the Regex filter accepts a custom pattern without surrounding slashes. Its Stop and reply action stops matching content and sends the configured response. Send fixed response lets you review that text before using it.

Code execution

Open Blueprint → Sandbox. Code execution exposes command tools to the agent. Turn it on only when the agent needs to run code.

After the draft saves, reset the Test Console and send this from the Messenger’s composer:

Use code execution to run python -c "print(2 + 3)".
Report the actual command output.

Expand the command’s tool card. A successful test shows Completed and a result of 5. A failed start shows Error with the failure detail. An assistant’s explanation alone doesn’t prove execution succeeded.

The sandbox also shows execution and session timeouts, network egress and prelude commands. Leave these unchanged when checking the first command. If a required file mount fails, the command cannot run. Resolve the error before relying on its result.