// documentation
Documentation
Everything you need to benchmark, guard, and certify your AI models. Start with the quick start guide or jump to a specific section.
Getting Started
Create an account
Sign up for free no credit card required. You receive 50 benchmark runs per month immediately.
Connect your models
Add API keys for OpenAI, Anthropic, Google, or any OpenRouter-compatible model on the Integrations page.
Define guardrail policies
Set system instructions, forbidden keywords, and expected output formats in the Guardrails section.
Run a benchmark
Select models, write a test prompt, and launch. Results are stored and tracked over time automatically.
Build your trust matrix
After 1,000 consistent, violation-free runs, your model earns a liwaiwai labs Reliability Certificate.
Benchmarking
// example: launch a benchmark run
POST /api/benchmark/run
{
"models": ["openai/gpt-4o", "anthropic/claude-3-5-sonnet"],
"prompt": "Summarise the following contract clause...",
"iterations": 100,
"guardrail_policy_id": "policy_abc123"
}Benchmark runs execute your prompt against all selected models concurrently. Each result is stored with latency, token counts, and guardrail pass/fail status. Schedule recurring runs to track consistency longitudinally.
Guardrails
Keyword rules
Block or flag responses containing forbidden terms legal, clinical, or sensitive language.
Format rules
Require JSON, markdown, or structured output. Flag any response that does not conform.
Semantic rules
Embedding-based checks to catch off-topic, hallucinated, or misaligned responses.
Full API reference and SDK documentation coming soon.
The platform is open. Start exploring immediately.