Console.
A live workbench for the doubt layer. Pick a policy, load a hard case or paste your own state, and read the verdict the same way your code would. Every answer carries a calibrated confidence, and thresholds decide what acts alone. See docs for the API behind this screen.
support inbound
Use it to Route tickets, emails, and messages. Best for support ops triaging mixed-language queues.
Read it as Act means auto-route. Review means a human should glance first. Escalate means low confidence or churn risk.
llm firewall
Use it to Screen prompts before they reach your model. Best for chat products, copilots, and agents with tool access.
Read it as Allow means clean. Review means suspicious. Block means jailbreak, injection, or leak risk at or above your threshold.
model router
Use it to Spend frontier-model money only where it matters. Best for gateways sitting in front of two model tiers.
Read it as Small means cheap and safe. Frontier means complex or high stakes. Human means do not automate this one.
content safety
Use it to Score user posts and uploads. Best for communities, marketplaces, and UGC feeds.
Read it as Allow means publish. Review means queue for a moderator. Block means toxic or threatening content.
Each loads a full record plus the options it needs. Hover a chip for why it is tricky. Press ⌘↵ to send.
A JSON object of fields, or plain text (sent as-is).
e.g. {"model":"multilingual","auto_act_above":0.9} · omit to auto-route by language.