Conversation analytics
Real-time, per-turn risk detection for character conversations.
API · SandboxSOLUTIONS / GUARDRAIL
Guardrail remembers safety signals about each end user across sessions, such as signs that the user may be a minor, and checks both user turns and character replies against them and your project policy. It returns a decision, guidance for your character model and, for replies, a safe rewrite.
Preview · SandboxWHAT IT DOES
Signals such as a likely minor are kept per end user, with their level and how long ago they were seen, and inform every later check.
Check the user’s turn before your model answers, and the character’s reply before it is sent.
Record age verification, a self-reported age, a moderator note or a risk signal from your own systems.
One call deletes everything remembered for an end user: signals, timeline and decision log.
A per-message filter forgets. A user who said they were fourteen yesterday is still fourteen today, so the safety layer has to remember too.
INPUT / OUTPUT
User turn or candidate reply · subject id · your own safety signals
Check against remembered signals and the project policy
Decision · guidance · safe rewrite · safety profile
Call it over REST or MCP with one API key. Sandbox keys are never billed; the full request and response reference, with the OpenAPI specification, is in the developer portal docs after you sign up.
MORE SERVICES
Real-time, per-turn risk detection for character conversations.
API · SandboxChat with a character that remembers, stays in character and stays safe.
API · SandboxLET’S BUILD WHAT’S NEXT
For technology partnerships, product integrations, and investment conversations.
Talk to Tridot