Add AI review to your application.
Submit a question to a saved panel and retrieve the review through the API. Use the responses and final conclusion in your own application or workflow.
API access is included in paid plans
Keep review configuration separate from each request.
An assembly stores the models, format, and settings. A debate request supplies the question and assembly ID. Reuse the configuration when you want a consistent review process.
Reviews run asynchronously. Keep the returned session ID, check status, and retrieve the result when it is complete. Handle failures and usage limits in your application.
What to look for in the review.
Reuse a configuration
Save the panel once and send different questions to it.
Retrieve the record
Access the model responses and final answer rather than only a summary.
Integrate at your pace
Start with the API quickstart, then add the operational handling your application needs.
Build the orchestration layer or call the review API
A framework may be right when orchestration itself is your product. The API is useful when you want the review behavior without owning every operational edge case.
| Engineering concern | DIY model chain | RigorDesk API |
|---|---|---|
| Panel configuration | Application code and prompt files | Saved Assembly resource |
| Duplicate protection | Custom request ledger | Required idempotency key |
| Long-running execution | Queue and worker implementation | Durable debate session |
| Completion delivery | Polling or webhook service to build | Status endpoints and webhooks |
| Budget visibility | Provider-specific accounting | Recorded usage and cost trail |
| Review protocol | Custom agent graph | Seven first-class formats |
| Audit output | Logs assembled after the fact | Transcript and chairman conclusion by design |
Prepare a useful review.
Give reviewers a clear question and enough context to examine it.
Pre-publication claim review
Send a generated article, report, or customer answer through a claim-focused Assembly.
Architecture and change approval
Trigger a Red Team panel when an RFC or high-risk change reaches review.
Design for failure, cost, and human escalation from the first call.
A multi-model run is slower and more expensive than a single completion. Set application-level timeouts around user expectations, not around wishful synchronous behavior. Use idempotency on every start request, store the returned session identifier, handle completed and failed terminal states, and make webhook processing idempotent too.
Do not turn the chairman into an automatic authority for consequential actions. Define thresholds that route uncertain, contradictory, or high-impact cases to a person. Protect API keys, minimize sensitive payloads, and document which downstream action is allowed from an AI conclusion versus which requires accountable approval.
What to know before you run the panel
What is the basic multi-LLM API workflow?
Create or select an Assembly, start a debate with an idempotency key and your case, store the returned session ID, then poll the debate or receive a webhook at completion. Read the final conclusion together with its transcript and usage data.
Should I use the API for real-time chat?
Usually no. Structured debate prioritizes review depth and an inspectable artifact over conversational latency. Use a normal model call for low-risk interactive turns and invoke the debate API at deliberate review gates where several minutes of scrutiny are justified.
Does agreement mean the answer is correct?
No. Models can repeat the same mistake or rely on the same weak source. Read the objections and check important claims against evidence.
Why use other models instead of asking one to critique itself?
Self-critique can help. Other models and specific review roles provide additional responses to compare. The format determines whether reviewers begin independently or see earlier answers.
Does this replace human review?
No. Use the findings to improve a draft or prepare questions for a responsible reviewer. The models do not approve a decision or verify every claim.
Start with an API request.
Follow the quickstart to create a saved panel, submit a question, and retrieve the result. API access requires a paid subscription.