Skip to content
Okashic Docs

Security & privacy

Okashic is built so that the easy path is the secure path. This page states what we guarantee, how we hold it, and — just as plainly — what we don’t.

Your vault is yours

Okashic encrypts your vault end to end. The keys are derived and unlocked only on your devices; our servers hold ciphertext they cannot read. A breach of our cloud exposes scrambled bytes, not your passwords or identity.

Agents get results, never your secrets

When you let an agent act for you, it never receives a raw credential. The local broker injects the secret into the request, filters the response down to the fields the task needs, and records the action in an audit log kept on your device. A hijacked or misled agent cannot turn access into a leak.

What we guarantee

These are enforced in the product today.

  • End-to-end encryption. The cloud stores ciphertext it can’t read.
  • Secure secrets. Not in a response, an error, a log line, or the audit log.
  • Filtered responses. On API connections your agent receives only the fields the task needs.
  • An audit log on your device. Append-only, so you can always see what was accessed, by which agent, and when.
  • Access you can withdraw. Revoke an agent or remove a connection at any time. It takes effect immediately.

What we don’t guarantee

Okashic can’t guarantee that an agent uses its access wisely. Once you connect a service, a misled agent could still take an action it’s allowed to take, and nothing interrupts that action mid-flight — the one-time connection approval is the gate. We bound that with the filtered, deterministic action surface, the audit log that shows what happened, and access you can withdraw at any time. You’ll see it in the log, and you can take the access back.

We try to break it ourselves

Every security-relevant change goes through an adversarial review: automated checks — dependency and supply-chain scanning, fuzzing of the parsers that read untrusted input, static analysis, and secret scanning — plus a structured attacker-mindset review of the cryptography, the broker boundary, and account isolation, where one pass looks for flaws and a second tries to refute them. Findings and their fixes are tracked, and each becomes a permanent regression test.

Our first such review found no critical or high issue. The three minor issues it raised are fixed and covered by regression tests.

Responsible disclosure

If you find a security issue, email security@okashic.io.