1. Separate keys for each conversation
Aether uses a different encryption key for each conversation. The example below shows how that key is derived from a master secret and a session ID.
// Dynamic 256-bit Session Key Derivation
function deriveChatKey(chatId) {
return crypto.createHash('sha256')
.update(process.env.ENCRYPTION_KEY + ':' + chatId)
.digest();
}
// Authenticated AES-256-GCM Encryption
function encryptMessage(text, chatId) {
const iv = crypto.randomBytes(12); // 96-bit unique IV
const key = deriveChatKey(chatId);
const cipher = crypto.createCipheriv('aes-256-gcm', key, iv);
let encrypted = cipher.update(text, 'utf8', 'hex');
encrypted += cipher.final('hex');
const authTag = cipher.getAuthTag().toString('hex');
return `${iv.toString('hex')}:${authTag}:${encrypted}`;
}
Random initialization values and authentication checks help detect changes to encrypted data.
2. Encrypted database connections
Aether OS stores data in Neon PostgreSQL. Connections use TLS 1.3 and verify certificates:
- Connection Pool Isolation: Backend services maintain parameterized query boundaries, preventing SQL injection vectors.
- Hardware Security Shields: Database backups, transaction write-ahead logs (WAL), and storage layers are encrypted at rest with hardware AES-256.
- Zero Plaintext Storage: Even in the event of unauthorized database inspection, message contents and agent outputs cannot be read without the temporary key used to decrypt them.
3. Checking streamed output
The streaming layer checks model output before sending it to the browser. It removes internal reasoning tags and filters content that should stay private.
How customer data is used
Customer chats and code are not sold or used to train foundation models. Access to project data is controlled through user credentials.
4. Separate agent workspaces
Aether Team separates agent workspaces. Provider API keys are stored in an encrypted vault and decrypted in memory when needed for a request.