Run a Red Team campaign
Respan Red Team sends adversarial messages to an agent and evaluates its replies.
Only assess an agent and its connected systems when you own them or have explicit authorization to test them.
Choose a connection method
In the Respan platform, open Red teaming → Campaigns, then select New campaign.
Connect an agent
This method supports local, private, stateful, and custom-auth agents. It requires Python 3.11+ and a Respan API key.

Keep the adapter process available until the campaign completes. The adapter should create a fresh conversation for each attack and return assistant text as a string.
Connect an endpoint
Use this method only when the target accepts OpenAI-compatible Chat Completions requests at a browser-reachable HTTPS URL.

Enter:
- A recognizable target name
- The HTTPS base URL
- A model name, when the endpoint requires one
- Any required headers as JSON
Before launching, verify CORS, firewall or VPN access, authentication, and the response shape. Assistant text must be available at choices[0].message.content. Use a narrowly scoped credential and keep the platform tab open while the campaign runs.
Use the sample sandbox
Choose ShopBot (naive) to explore the workflow without sending requests to your systems. The target is intentionally weak and does not require target credentials.

Review the authorization acknowledgement, then select Launch campaign.
Authorize and launch
Before starting any target:
- Confirm the assessment scope and permission.
- Prefer a non-production environment.
- Limit the connected agent’s credentials, network access, and tool permissions.
- Check the campaign allowance and reset date shown in the creation screen.
Queued and running campaigns reserve allowance. Failed campaigns do not count as completed campaign usage. If no allowance remains, wait for the displayed reset date or change the organization’s plan.
Run from CI
Set RESPAN_API_KEY in the CI secret store, then apply a grade threshold:
Exit code 4 means the result fell below --fail-under.