Skip to main content
Every integration connects the same way. The specifics differ — which provider, which scopes, which account — but the shape does not.

Authorise

You are redirected to the provider, sign in there, and approve the connection. Tied never sees the credentials you type — the provider hands back a token instead.Sign in as the account you actually want the agent acting as. This is the most common thing to get wrong, and the consequences are invisible until something fails or, worse, succeeds under the wrong identity.

Grant scopes

Choose what Tied may do. This is the decision that matters.Grant the narrowest scope that does the job. If your agent only needs to look up orders, grant read. You can widen later; a scope you granted and forgot is a scope that exists forever.

Verify

Confirm the connection works before relying on it — run a real lookup against real data and check the answer is right.A connection that authorises successfully but returns nothing useful is a common failure, usually because the authorising user lacks access to the data your agent needs.

​
Choosing scope

Read and write are not a spectrum, they are a threshold. Read access means the worst case is your agent tells someone something it should not have. Write access means the worst case is your agent changes something in a system of record.Both are manageable. They are not the same risk, and they should not be the same decision.
Practical sequence that works well:
  1. Connect read-only first and let the agent run on real traffic.
  2. Watch what it actually does — which lookups, how often, whether the answers are right.
  3. Widen to write only where you have evidence the agent is behaving well on reads.
On Advanced, writes come with narration and rollback, which makes step three considerably less frightening than it sounds.

​
What the agent does with it

Once connected, the integration’s actions become available to the agent, which decides when to use them based on the conversation. You do not wire specific actions to specific questions — that is what the agent is for. If you need a fixed sequence, that is a chatbot flow instead.
A flow cannot call an integration directly. There are no webhook or HTTP-fetch nodes in the flow engine. Anything that reads from or writes to an external system is the agent’s job. See Node types.

​
Disconnecting

Disconnecting revokes Tied’s access immediately. The agent loses those actions, and conversations that depended on them will start escalating rather than resolving — which is the correct failure mode, but worth expecting. The action audit trail is retained. Past resolutions remain provable from their receipts even after the integration is gone.

​
Next

AI agents

What the agent does with what you connected.

Resolution pricing

How scope determines which resolution classes you can reach.

Plans

Which tier grants read versus write scope.

Setup checklist

Where this fits in the order of setting Tied up.