Overview
Vault keeps your agent's secrets — API keys, tokens, env vars — out of two places: the Clawise server (which only ever sees ciphertext) and the LLM provider (which only ever sees clawise://<id> placeholders).
Why Clawise Vault
| Tool | Risk |
|---|---|
.env file / env vars | Pasted into LLM prompts the moment your agent reasons about them |
| 1Password / Bitwarden | Decrypted into the agent's process memory, then into prompts |
| HashiCorp Vault, Doppler | Same — central server fetches plaintext, app uses it directly |
| Clawise Vault | Encrypted on your device; only the target service ever sees plaintext |
Traditional secret managers protect secrets from outside attackers. Vault protects them from everything in your agent's request path:
- The third-party LLM
- Proxies and aggregators in between
- Request logs along the way
- Prompt-injection attacks that try to make your agent exfiltrate a key through a tool call
Architecture
Clawise server
encrypted blobYour
agent
sk_live_xxx9zKmsk_live_xxx9zKmClawise daemon
master keyclawise://a3f7b2c4clawise://a3f7b2c4LLM
- Encrypted storage — credentials are AES-256-GCM encrypted on your device with your master key. Only ciphertext is uploaded; the server has no master key and cannot decrypt anything.
- Local daemon proxy — outbound LLM requests pass through the daemon, which strips real values and substitutes
clawise://<id>placeholders. Inbound responses get placeholders resolved back to plaintext before reaching your agent.