# A tool stopped working after connecting

It worked before governance and now returns an error. Almost always one of the things below.

## No provider key — error 424

The tool is in **managed API key** mode and there is no matching key. Claude Code needs an
Anthropic key; Cursor and Copilot need OpenAI.

Fix it on **Providers**, or move that tool to subscription mode in **Policy → Connection
modes**.

## Ambiguous provider key — error 409

Two keys of the same type both list the requested model in their enabled models. Sentilai
will not guess which one to bill. Make the model lists disjoint.

## Cursor needs its model selected

After connecting Cursor, open its **Settings → Models**, make sure the OpenAI key is
enabled and a `gpt-…` model is chosen. Otherwise Cursor quietly uses its own models and
chat may behave oddly.

## Copilot BYOK is not enabled on your GitHub organization

If VS Code does not offer a **Custom Endpoint** option at all, that is a GitHub
organization setting. Sentilai cannot enable it.

## An ungoverned agent got the device blocked

If the error reads *"access blocked: an ungoverned AI agent (…) was detected on this device
— remove it to restore access"*, nobody revoked anything. Your organization's
ungoverned-agent policy is set to block, and that machine has an autonomous AI agent on it.

The developer fixes this themselves: uninstall the agent, and access returns within about
half a minute of the next check. No ticket, no admin action. **Diagnostics** shows which
device and which agent; [Ungoverned-agent policy](/books/policy-and-risk-detection/page/ungoverned-agent-policy) explains the setting.

## It is a policy block, not an error

Check **Activity** and filter by outcome **Blocked**. A request stopped by your own policy
is not a malfunction. The Signals column names the rule that stopped it.