How does Palladin protect data?
Palladin lets AI agents use passwords and API keys under your control. Data is encrypted on your device. You decide what an agent can access and how it can use that data.
Zero-knowledge guarantee
Zero-knowledge means the server stores encrypted vault content without the keys needed to read it. Encryption and decryption happen in the user app or the agent’s local program.
The server still manages accounts, permissions, and operation records. API keys used to connect to Palladin are separate from secrets stored in a vault.
Who can access data?
- AI model
task intent - CLI / MCP
operation interface - Local Palladin program
local key use - Palladin API
permissions and encrypted data
| Participant | Responsibility |
|---|---|
| User client | Unlock, encryption, and access approval. |
| Backend | Identity, current authorization, limits, and ciphertext storage. |
| Native Agent runtime | Material verification, local decryption, and operation execution. |
| AI model | Requests a task; plaintext access depends on the chosen method. |
| Target process or site | Receives the data needed to use a credential; part of the disclosure boundary. |
Three kinds of control
Encryption protects content. The server checks permissions. The local Palladin program performs the allowed operation. Each layer has a different job. Blocking another retrieval does not erase data a recipient already copied.