Skip to the content
TEND is in beta.Limitations

See and control what your AI agents do

TEND checks what your AI agents try to do and holds risky actions until you say yes.

Create an account

Read-only, with sample agents. No sign-up.

13 integrations verified · 7 checks each

Dusk over forested hills

Interactive demo. Nothing real runs.

TEND

Overview

1 request needs your answer

Spends money

ops-agent wants to pay invoice 2026-118

Asked 2m ago · Expires in 8:00

The exact request

Pay invoice 2026-118: 4,800 EUR to Acme GmbH
Item
2026-118
Why you are asked
This action needs approval

Nothing happens until you allow it. Allowing lets this exact request run once. Change what needs your OK

Agent
ops-agent
Tool
pay_invoice · pay
Asked
2m ago
Matched rule
Payments wait for a person
Agents running
2
of 3
Waiting for you
1
right now
Allowed without asking you
14
last 24 hours
Blocked
2
last 24 hours

Works with

  • Claude Code
  • Anthropic
  • Model Context Protocol
  • LangChain
  • CrewAI
  • Ollama

AI agents now send mail, change records and move money. TEND puts a checkpoint in front of every tool call, and keeps a person in charge.

13
integrations verified in TEND’s own test
7
scenarios each of them passed: allow, deny, approve, reject, revoke, TEND unavailable and audit log
1
run per approval. A changed, reused or expired approval is refused.

One checkpoint for every AI agent

Connect the agents you run, decide what each may do, and answer what is held. Everything ends up on record.

Decide what runs

Held requests wait with their exact text. An approval works once.

Prove what happened

Every request and decision, chained to the one before it.

An agent asks.

Pick what the agent asks for. TEND checks it against the rules before anything runs.

What the agent asks for

The rules

  • read_ticket.readAllow
  • send_email.sendAsk me first
  • pay_invoice.payAsk me first
  • delete_record.deleteNever
  • shell.runAsk me first
  • *Blocked

Pick what the agent asks for. TEND checks it against the rules before anything runs.

Set your own rules

Everything you need to stay in charge of your agents

Swipe to explore

  • Approvals

    Held requests wait with their exact text. Allow once, or don’t.

  • Rules

    Allow, Ask me first or Never, per tool. Anything not listed is blocked.

  • Activity

    Every request and decision, hash-chained to the one before.

  • Agents

    Each agent has its own key. Revoke one, the others keep running.

  • Checks

    One claim per check: Verified, At Risk or Not Verified.

  • Gateway

    TEND holds the credential and runs the tool itself, once.

  • Command line

    See what waits, approve and read the log from your terminal.

  • Integrations

    Hooks, adapters and an MCP proxy for the agents you already run.

Clear about what it controls, and what it doesn’t

TEND is in beta. Every page says what it does not do, and a status only comes from a test that really ran.

Security limitations
  • Not a sandbox

    TEND covers only the calls that pass through it. An agent that skips it, or holds the credentials itself, can act without TEND.

  • No threat detection

    No content scanning and no prompt-injection detection. TEND can’t judge what a request means.

  • Not independently audited yet

    TEND is in beta and has had no independent security review.

Every tool call your agents make.Checked first.

Create an account
Dusk over forested hills