$UMBRAI is live. CA: CjSLVVpKdzqogxmWcAdJrxNb1CQAq9Fb3ea84YsYpump Encrypted AI research on Solana. Your prompts are encrypted before they leave your browser. The server never sees plaintext. No account. Pay per query. Withdraw anytime. Follow 𝕏: x.com/UmbraIntelApp Website: umbraintel.xyz/ Github: github.com/domcomit/umbra-sd…

The SDK is open source and published on npm. You can read exactly how the encryption works before trusting it with a single query. The ECDH key exchange, the AES-256-GCM implementation, the envelope format — all of it is in the repository. Privacy tools that can't be audited are just promises with better branding. npm install @umbra_intel/sdk github.com/domcomit/umbra-sd…

Every Umbra session gives you a transaction hash. Open the Verify terminal on the landing page, paste the hash, and see exactly what was published to Solana: session ID, parameter hash, timestamp, wallet address. Never the prompt. Never the response. Or skip the copy-paste - every chat answer has a "Verify on-chain" link built in. One click, opens the terminal pre-filled. This is what accountability looks like before the enclave launches. The record is minimal by design, public by default, and verifiable by anyone without asking Umbra for permission.

Every session gets its own encryption key. One session, one key, then it's gone. This is called forward secrecy. If someone somehow obtained a key from a previous session, they still couldn't decrypt any other session - because that key was never reused and no longer exists. Your browser generates a fresh key pair every time you open a chat. It computes a shared secret with the server, encrypts your prompt, and the key never leaves your device in a recoverable form. Past sessions can't be retroactively broken. That's not a UI feature. That's cryptography.

The people who need this most aren't privacy researchers. They're the founder who just asked ChatGPT to analyze why their churn is spiking — and disclosed their retention problem to OpenAI's training pipeline in the process. The analyst who ran a thesis on a public AI tool three days before a trade. The BD lead who used Claude to draft partnership terms and attached the deck. None of them thought of it as a data leak. It looked like a productivity tool. That's the problem. Umbra is for anyone who uses AI for anything that would hurt if a competitor saw the question.

ChatGPT Enterprise costs $30+ per user per month and promises your prompts won't train the model. It doesn't promise nobody reads them. It doesn't promise hardware isolation. It doesn't give you a cryptographic proof that your session was private. You get a contract clause and a trust-me. Umbra gives you a different answer: the server receives ciphertext. It is technically unable to read your prompt in plaintext - not by policy, by architecture. One is a privacy promise. The other is a privacy proof.

Every session gets its own encryption key. One session, one key, then it's gone. This is called forward secrecy. If someone somehow obtained a key from a previous session, they still couldn't decrypt any other session - because that key was never reused and no longer exists. Your browser generates a fresh key pair every time you open a chat. It computes a shared secret with the server, encrypts your prompt, and the key never leaves your device in a recoverable form. Past sessions can't be retroactively broken. That's not a UI feature. That's cryptography.

Updated CA on Github:

The server stores no conversation history. Zero. When you close the chat tab, the session is gone from the server's side. Your history lives in your browser's local storage - under your control, not ours. Clear it and it's gone permanently. Most AI products treat your history as a feature. We treat it as a liability we don't want to hold. No conversation database. No session logs. Nothing to subpoena, nothing to breach, nothing to sell.

For developers: Umbra is a three-line integration. Replace your OpenAI endpoint with the Umbra endpoint. Sign requests with a wallet. Pay per call. Your existing AI code works. Your users get privacy. You get revenue share on queries. Also shipping an MCP server - private AI with encrypted web search and on-chain data, all inside the enclave boundary.

Dossier Mode is the one I'm most excited about. You name a company or a market. Umbra runs a full automated research session – scrapes the web, synthesizes data, finds gaps, writes a strategic summary. Delivers it as an encrypted PDF. The entire research chain is private. The AI providers involved each see one small piece. Nobody sees the full picture except you.

Before your question even leaves your browser, Umbra encrypts it. The server never sees what you typed. Not Umbra's server. Not the AI provider's server. Nobody. The encryption key comes from your Solana wallet signature – nothing stored, nothing to steal. You've heard of end-to-end encrypted chat. This is end-to-end encrypted thinking.

But someone still needs to decrypt the question to get an answer, right? Yes. And that's where it gets interesting. Decryption happens inside a hardware-isolated chip called a Trusted Execution Environment. AWS Nitro Enclave. It's physically sealed – even AWS engineers can't see what's running inside. The enclave decrypts your question, sends it to GPT or Claude, gets the answer, re-encrypts it, sends it back. Then destroys everything. No logs. No storage.

Umbra Chat is the main product. Think ChatGPT but for competitive research – and you don't leave any footprints. Built-in templates for competitor analysis, market sizing, pricing intelligence, and SWOT breakdowns. Connect wallet - fund with $UMBRAI - research - done. No history saved anywhere. Not even on your own device unless you choose to keep it.

Every dev fee Umbra collects goes to a single wallet. That wallet is publicly linked to our GitHub account. 4JG6tiMgRZHz14yWC79sMdCNQXv2UsP6BFtc592WwozJ github.com/domcomit/umbra-sd…