LLM Playground

…

Send a prompt through Vouch's Trust Verification Layer — every response gets an Ed25519-signed certificate.

Demo — limited threat coverage. Vouch v0.26.0 ships with classifiers for technical attacks: prompt injection, PII, cipher/encoding, model extraction, adversarial suffix, semantic drift, and output anomaly. It does NOT yet ship with harmful-content / criminal-intent / violence / self-harm classifiers — those are queued for the next release. Refusals from the upstream LLM are PASS-able by design; harmful generations are not automatically flagged in this build. Do not use this playground as evidence Vouch covers content-safety risk.

Model

Claude (Anthropic) no key → mock
claude-opus-4-5 claude-sonnet-4-5 claude-3-5-haiku
OpenAI no key → mock
gpt-4o gpt-4o-mini o4-mini
Grok (xAI) no key → mock
grok-3 grok-3-mini grok-2-1212
Perplexity no key → mock
sonar sonar-pro sonar-reasoning
Mock (offline, no key needed)
mock-llm

Auth

Prompt

Results appear here after you send.

Perplexity is pre-selected and ready to use live.