> For the complete documentation index, see [llms.txt](https://docs.slinky.network/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.slinky.network/core-concepts/how-requests-are-routed.md).

# How requests are routed

The chat and the tool network serve different purposes. The chat holds the conversation. A downstream tool should receive a specific job that it can complete, not every message that led to that job.

For example, a market-data request for the current ETH price needs an asset, quote currency and time window. It does not need earlier discussion about another asset, a private investment thesis or a wallet label.<br>

<div data-with-frame="true"><figure><img src="/files/8cYCn9oKIFwh5W5D9cie" alt=""><figcaption></figcaption></figure></div>

{% hint style="info" %}
The routing step keeps the fields needed for the job and excludes unrelated context.
{% endhint %}

#### What the router does

1. Identifies the capability needed for the next part of the task.
2. Extracts the fields and instructions relevant to that capability.
3. Excludes unrelated history and identity labels by default.
4. Sends the prepared request through the Tor egress path.
5. Returns the result to the original conversation.

This is selective disclosure, not automatic removal of every sensitive fact. If a user asks a tool to investigate a named person or wallet, that name or address is part of the task and the tool will receive it.