What Lev does
Lev - AI assistant - helps find published material, structure a problem, create an initial plan, explain a topic, and prepare a safe next step. Lev is not a person and may be wrong.
Data it relies on
The assistant uses OpenAI for the conversation, but searches only PUBLIC or FEATURED records in the canonical local Supabase/PostgreSQL knowledge base before making factual claims. Semantic indexing runs locally with BGE-M3, while Neo4j stores a separate graph projection. If no verified material exists, it must say so plainly.
Your messages
Text is used to answer and, with your consent, to pass relevant context to a person. Do not send secrets, passwords, codes, exact account numbers, identity documents, or third-party data.
Persistent personalized memory, use of conversations for system training, and marketing require separate consent.
Voice conversations
ElevenLabs only recognises speech and reads an approved response aloud. The recognised text enters the same mandatory route as a typed message: Luna checks safety and intent, selects any permitted skills and MCP connections for Sol when needed, and verifies the final answer.
No separate voice agent answers the user or connects directly to the knowledge base. Voice therefore cannot bypass source rules, financial boundaries, prompt-injection protection, or final verification. The site does not retain the audio file.
Financial boundaries
The assistant may provide general information about principles, risks, portfolio structure, and scenarios. It does not provide formal individualized investment advice or execute transactions.
Actions and confirmation
- Lev may prepare a consultation request from relevant conversation context but cannot submit it without your explicit confirmation;
- for a signed-in user, name and verified email are securely taken from that user’s own account, so Lev does not ask for them again;
- after both consents are confirmed, the request is stored in Supabase/PostgreSQL, appears in Studio, and is delivered to the responsible person;
- every machine client has a separate revocable token and limited scopes;
- financial transactions, booking a specific calendar slot, bulk email, and publishing are not performed by the public agent.
Quality and control
Critical flows are tested for factuality, source use, privacy, prompt injection, safe refusal, and language quality. Errors can be reported through the contact form.
Two ways to interact
Start with Lev to find material and frame the problem, or go directly to a service, read about Mykhailo Patsan, and submit a consultation request. A person responds to submitted requests.