Skip to content
Okashic Docs

Connections

A connection is how an agent uses one of your services without ever seeing the credential. You approve it once, in Okashic. After that the agent can use the service, and every action it takes is recorded.

Agents

An agent is software that acts for you — like Claude or Hermes. Okashic doesn’t build agents; it’s the layer they read from and act through.

Each agent identifies itself to Okashic with its own API key. You create, view, and revoke keys yourself. Revoking a key cuts that agent off immediately.

Connections and kinds

You create a connection once, in the Okashic app, never in a chat. It’s a one-time approval: you approve a service a single time, and the agent then works without asking again. An agent can request a connection, but only you can create one, so a poisoned page or email can make an agent ask, but can’t give itself access.

When you add one, Okashic shows the kinds you can set up:

  • Website sign-in — a site you log into. The broker signs in and holds the session for your agent.
  • API token and OAuth — a service you reach with a token or an OAuth sign-in, like Google or GitHub. The broker calls it and filters the response down to the fields the task needs.
  • No credential — a public API that needs no secret.
  • Custom (REST) — any other REST API you define yourself.

Whichever kind, the broker applies the credential in-service and hands the agent only the result. The password, token, or session cookie never reaches the agent.

Who can use a connection

By default a connection is open to all the agents you’ve set up. You can narrow it on the connection itself: limit it to the agents you choose, and on an API connection pick which operations each agent may call. An agent sees only the connections it’s allowed to use, so a restricted connection stays invisible to the rest.

Sessions

A website sign-in connection is the broker holding a live, signed-in session on your behalf. It’s how an agent acts on a site that has no agent-friendly API. The agent can act within that session, so its reach is the session itself, bounded by the audit log and by your ability to remove the connection at any time.

Audit

Every action an agent takes is logged on your device: which agent acted, what it did, when, and the outcome. The log is append-only and produced on the device, because the server can’t read your vault.

You can review, search, and filter it to answer what an agent accessed and when. The local history is always there on your device.

Withdrawing access

Remove a connection, disable or revoke an agent, or delete a record at any time. The change syncs and takes effect immediately, so a mistake stays bounded and recoverable.