Skip to main content
An endpoint is the authenticated address your audience connects to. One endpoint serves AI agents (Claude, ChatGPT, Cursor) over MCP and your own apps through the SDK with an API key. It searches one or more of your sources and answers every request through a single search tool. Each endpoint has:
  • A sign-in provider. Your users sign in before they can search. There is no open, no-auth option.
  • One or more sources. When an endpoint searches several sources, Raily merges the results into one list and balances them across the sources.
  • Its own search config. Search behavior, rules, card display, the tool description, and version history all live on the endpoint. Changing them never changes your sources or your other endpoints.
  • Analytics. Raily records every call, so you can see who searched what and how fast the endpoint answered.
New endpoint form with name, collections, who can connect, and Search results with UI

Create a new endpoint

Create an endpoint

Your endpoints live in the Your audience panel on the right of Search Rail. Each one shows the sources it searches and whether it’s active. Click New endpoint to add one. You can also start from a source: in its Searched by row, click add endpoint, then pick an existing endpoint or choose Create a new endpoint for that collection. In trial mode, Raily has already created one endpoint for you. You can’t add a second, but you can point it at any collection you connect. Before you begin, connect a source and let Raily finish setting it up, and make sure the Organization has a sign-in provider. The form tells you if either is missing.
1

Name the endpoint

Type a Name your audience recognizes (for example partner-api). Raily assigns the endpoint URL when you click Create endpoint.
2

Choose the collections

Under Collections, click Choose (or Change if you started from a source) and pick one collection or several. Each one appears as a chip; click its × to drop it. With more than one selected, the form confirms that the endpoint searches them all through one search tool. Searching several sources from one endpoint is available on paid plans.
3

Choose who can connect

Pick a sign-in provider under Who can connect. Your users sign in with it before searching.
4

Set how results render (optional)

Leave Search results with UI on and your users get a rich, interactive results page inside Claude or ChatGPT, branded with your data. Turn it off for plain text. More in Search with UI below.
5

Add internal notes (optional)

Open Advanced for a Description (an internal note for your team, never shown to agents), Agent instructions (sent to agents when they connect, to tell them how to use this endpoint), and rate limits.
6

Create the endpoint

Click Create endpoint. Nothing is saved before this.
The new endpoint starts from the search config Raily already generated for each source, so it answers right away. From then on the settings belong to the endpoint: edits you make on it stay on it.

Add or remove sources

Open the endpoint and find What this endpoint serves.
1

Open the source picker

Click Manage sources. Every source the endpoint searches today is already selected.
2

Pick the sources

Select the sources to add and clear the ones to drop, then apply. An endpoint always keeps at least one source.
3

Wait for the live test

Raily updates the endpoint, runs a live test search, and rolls the change back if the search returns nothing. A message confirms which sources were added or removed.
To drop one source quickly, click the × on its chip. The × appears only when the endpoint searches more than one source. You can also add a source from its own page. In the Searched by row, click add endpoint and pick the endpoint. Raily asks you to confirm (Add it here). Nothing the endpoint already searches is removed. On a trial plan the button reads Change instead, and picking a collection switches the endpoint to it. When an endpoint’s sources change, Raily regenerates its search settings with AI to fit the new set. The Search Behavior tab shows Regenerating with AI… while it runs, then a note that names the sources added or removed and how many settings changed. Click Review changes to see them in History, or Restore previous to undo the regeneration.

Search settings

Open the endpoint and click Search config to change how its searches run: rules, search behavior, the reranker, and how result cards look. Each tab is described in Agentic Layer. The tool description and version history are covered below.

Write the tool description

Agents like Claude and ChatGPT read the tool description to decide when to search this endpoint. Edit it on the Tool description tab. The limit is 8,000 characters. One source. Agents read the description Raily generated for the endpoint’s source. Edit the text and click Save to use your own, or click Regenerate with AI to have AI write a new one from the source’s description. Several sources. Agents read one description combined from each source’s description. Raily combines it again whenever a source is added or removed, or a source’s description changes. Choose how it’s combined:
  • Combine with AI writes one description from all the parts. It takes about a minute, and agents keep reading the current description until it’s ready.
  • Combine using code joins the parts without AI. If a part has to be shortened to fit the limit, the tab tells you which source’s description is cut.
To use your own text for the whole endpoint, edit it and click Save, then confirm Save my description. Your text no longer updates when sources change. To go back to a combined description, click Regenerate with AI; your text stays in the endpoint’s history. Regenerating or combining with AI uses your AI generation budget.

Restore an earlier version

Raily saves a version each time the endpoint’s search config or its sources change. Open the endpoint and click Config history, or open the History tab on its search config page. Each version shows how it was made (Generated, Manual edit, or Sources changed) and when.
1

Compare a version

Click Diff vs live to see what differs from the current config.
2

Restore it

Click Restore and confirm. The version becomes this endpoint’s live search config right away. Its sources and your other endpoints don’t change, and the current config stays in history so you can switch back.
A version saved when the endpoint searched a different set of sources can be compared but not restored.

Search with UI

Search results with UI controls how this endpoint presents results in the connected AI client. Turn it on and the endpoint returns an interactive results page that the client renders in the chat: the MCP UI in clients like Claude, or the app UI in ChatGPT. Turn it off and the same results come back as plain text, which every client can read.
The Search with UI toggle on an AI Endpoint

Search with UI

With it on, a reader can scan, filter, and open results visually without leaving the chat. The fields shown on the cards come from your Data Catalog and the endpoint’s Display tab.

Analytics

Raily records every call to every endpoint:
  • Total Calls: the number of requests to the endpoint
  • Average Response Time: how long the endpoint took to answer
  • Total Input Tokens and Total Output Tokens: tokens sent in requests and returned in responses
  • Average Input/Output Tokens: per-request averages
  • Access patterns over time: a timeline of usage
See them in the Raily dashboard at app.raily.ai/dashboard.

Next steps

Providers

Add a sign-in provider for your endpoints

Analytics

View endpoint analytics and metrics