Skip to the content
TEND is in beta.Limitations

TEND / How it works

Every tool call stops at TEND first

TEND sits between your agent and the tools it uses. Before a tool runs, TEND checks the exact call against your rules: it lets it through, holds it for you, or blocks it. Anything not listed is blocked.

Four steps, every time

  1. Your agent asks

    Before a tool runs, the agent’s TEND wrapper, hook or MCP proxy sends TEND the exact call: the tool, the action and what it will be called with.

  2. Your rules decide

    Each tool is Allow, Ask me first or Never. A rule can look at the arguments, for example an email only to your own domain or a refund up to 50.

  3. You answer what is held

    A held request waits in Approvals with its exact text. Allow once lets that exact request run one time; a changed, reused or expired approval is refused.

  4. Everything is on record

    Requests, decisions and rule changes go to an append-only log. Each entry is chained to the one before it, so a changed or removed entry shows up.

Two ways to put TEND in front of a tool

TEND authorizes; your integration enforces. How strong the protection is depends on where the credential lives.
  • Checkpoint

    Your agent asks TEND before each action and acts only on a yes. Quick to add with TEND’s adapters, hooks and MCP proxy. It is cooperative: an agent that skips the check, or holds the credentials itself, can act without TEND.
  • Gateway

    TEND holds the credential and runs the tool itself, once, after the call is allowed. Your agent never holds the key, so it cannot go around TEND for that tool.

If anything goes wrong, nothing runs

If TEND can’t be reached, answers with an error or its database is unavailable, the protected action does not run. An unreadable rule set denies everything. Each allowed call runs at most once: a retry, or the same call delivered twice, is refused.

Put a checkpoint in front of your first agent