help
FAQ
Questions answered
Everything you need to know about using Lexi.
Billing
Billing
What happens if there's no saving?
Nothing. If STONE cannot reduce tokens on a request, the original is forwarded unchanged and there is no Lexi fee. You only pay the standard provider cost — the same as calling the provider directly.
When am I charged?
You are charged per request. Each response includes headers showing the exact cost breakdown: provider cost, Lexi fee, and tokens saved.
Do I get free credits?
Every new account receives $10 in free credit. No credit card is required to sign up.
How does Lexi charge me?
Lexi charges 40% of the token savings it creates. You also pay the provider cost for the tokens actually sent. The formula is:
X-Lexi-Request-Cost-Cents,
X-Lexi-Savings-Cents,
X-Lexi-Balance-Remaining . If the original is sent unchanged, there is no Lexi fee.Can my balance run out?
Yes. Your balance is prepaid credit. If it reaches zero, requests are rejected until you top up. You can check your balance in the dashboard or via the X-Lexi-Balance response header.
Security
Security & Privacy
Does Lexi need my provider API keys?
No. Lexi uses its own provider accounts. You authenticate with a Lexi API key. Your provider keys are never needed or stored.
Does Lexi store my conversations?
Lexi processes conversation context to restructure it. Session data is held in memory during the conversation and encrypted at rest (AES-256-GCM). Data is automatically purged after the session ends. Lexi never trains on your data.
Where is my data stored?
Lexi runs on Azure in Europe. All data is encrypted in transit (TLS 1.2+) and at rest (AES-256-GCM).
Technical
Technical
Does restructuring affect response quality?
In blind benchmarks against GPT-4o, Lexi scored 8.4/10 recall accuracy versus 9.0/10 for full context. The gap is in peripheral detail, not factual accuracy. If restructuring cannot help on a given request, the original is sent unchanged — quality never drops below baseline.
What about long conversations?
STONE uses bounded resources — token usage stays flat regardless of conversation length. In a 75-turn benchmark, STONE maintained 91.6% token reduction with 8.4/10 recall accuracy. The architecture is designed for conversations of any length.
What is STONE?
STONE (Semantic Token Optimization and Natural Encoding) restructures conversation context before each API call. It preserves meaning while reducing tokens — bounded resources, fact pinning, and cold recall ensure nothing important is lost.
Which models does Lexi support?
Lexi supports 33 models across OpenAI, Anthropic, Google, Mistral, xAI, DeepSeek, and Meta. See Supported Models.
Integration
Integration
How long does integration take?
About two minutes. Change your base URL and API key — no other code changes needed. See the Getting Started guide.
Does Lexi support streaming?
Yes. Lexi supports streaming responses for both OpenAI and Anthropic endpoints. STONE restructuring happens before the stream begins.
Does Lexi work with Anthropic's API?
Yes. Lexi supports the
/v1/chat/completions
endpoint for Anthropic's Messages API. Use your Lexi API key and point to /v1/messages .Still have questions?
Still have questions?
We're here to help.
Or start free and see for yourself. $10 credit, no card required.
$10
Free credit on signup
$0
No card required
2 min
2-minute integration
<1 day
< 1 business day response