Frequently Asked Questions

What does Souma do today?

Souma is the output compliance layer for regulated AI teams. The current release is centered on right-to-use review: submit URLs and usage purpose, receive a verdict, then review evidence in Pulse, Compliance, Evidence, and Checker.

How do users access the product now?

Access starts from the homepage drawer:

• Login for existing workspaces
• Create New Workspace for new teams

There is no separate standalone login page anymore.

How is Souma different from guardrails tools?

Most guardrail products focus on prompts or safety filters before generation. Souma is aimed at the output release point, where a team needs to know whether generated work can actually be used, shipped, or escalated.

What AI models does Souma support?

Souma is model-agnostic. It works with OpenAI, Anthropic, Google, Mistral, open-source models, and custom fine-tuned deployments. One integration covers your entire AI stack.

How should we handle large RtU batches?

Keep them asynchronous. The current checker uses small batches and limited concurrency so the UI does not collapse under timeouts. That same approach is recommended for production integrations:

• chunk the workload
• store every validation_id
• stream or poll for completion
• retry only failed batches

What does the Evidence view store?

Evidence is the operational record for a validation:

• URL
• usage purpose
• verdict
• confidence score
• reasoning
• conditions
• timestamps
• policy references

Is provenance a separate live workflow yet?

Not fully. The current live product is strongest in RtU operations. Provenance-related data already shows up in evidence and workflow configuration, but the dedicated provenance experience is still expanding.

Where do we start for procurement or deeper review?

Use the workspace request flow or commercial contact path to start product, legal, or security review. The docs focus on the product workflow; procurement-specific materials are still handled through the team directly.