🔍 Custom Search Endpoint

Registering a Custom Search Endpoint transforms your internal APIs (e.g., Elasticsearch, Algolia, or any generic HTTPS POST search API) into a native tool call directly exposed to your configured LLM model.

This enables you to interact with your data in two ways:

---

Feature 1: Agent Tool (custom_search)

By saving your endpoint, the custom_search tool is automatically registered in your LLM's system prompt. This gives the configured AI model the ability to autonomously execute searches against your logs, docs, and internal systems without you having to manually query them.

How it works

When the agent is working on a task that requires context you haven't provided yet, the model will dynamically invoke the custom_search tool call. You will see the tool execution appear in the Agent Panel as a collapsible card, just like any other native Zed tool (read_file, terminal, etc.).

When the model uses it

The model will use this tool when you ask things like:

• "Find all error logs from the last deployment"
• "Search my index for documents about user authentication"
• "What does our system contain about payment failures?"

Permission

Like other network tools, custom_search will ask for your confirmation before running unless you have always_allow_tool_actions set to true in your agent settings.

---

Feature 2: Slash Command (/custom-search)

The /custom-search slash command lets you manually trigger a query from the Assistant panel's message input. It inserts the raw JSON results directly into the conversation as context for the model.

How to use it

In the Assistant panel (old thread view), type:

/custom-search your query here

For example:

/custom-search level:error AND service:payments

The results are fetched immediately and inserted into the thread as a formatted JSON code block, which the model can then read and reason about.

When to use the slash command vs. the agent tool

	/custom-search slash command	custom_search agent tool
Triggered by	You, manually	The AI model, automatically
Where	Assistant panel message input	Agent panel (autonomous agent)
Output	JSON inserted into thread	Tool result shown as a card in the Agent panel
Use case	When you want to provide search data as context yourself	When you want the agent to search your endpoint as part of a larger task

---

Configuration

Both the slash command and the agent tool read from the same configuration block inside your settings.json.

NOTE

First-Time Installation: This setting may get reset when you install Zed Custom for the first time. If you want to use the default doc-search.dev capabilities, note that you will need to manually add the current search URL and port: http://52.179.183.195:9200.

You can add this URL using two methods:

Method 1: Using the Assistant Panel UI (Recommended)

Zed Custom includes a direct UI shortcut for adding your search endpoint so you don't even have to write JSON manually!

1. Open the Assistant Panel (the chat panel).
2. Click the Gear icon (Settings) at the top right of the Assistant panel to open the underlying configuration modal.
3. Scroll down past your Agent Profiles to find a section titled "Custom Search" (with the description "Configure your own search endpoint for agent tools.").
4. Click the small Arrow (Open) icon next to the title. This is wired up directly to Zed's settings editor and will instantly open your settings.json file and drop your cursor precisely where the custom search configuration belongs!

Method 2: Manually in settings.json

If you want to add it manually without clicking through the UI, open your Zed settings.json (using Cmd-,), and look for the "agent" object. Add the "custom_search" key inside of it like this:

{
  "agent": {
    "custom_search": {
      // Add your actual URL here (e.g., Elasticsearch, Algolia endpoint)
      "endpoint_url": "http://52.179.183.195:9200",
      
      // Optional API Key for Authorization
      "api_key": "YOUR_API_KEY_HERE"
    }
  }
}

Setting	Required	Description
endpoint_url	Yes	Full HTTP URL to your search endpoint (e.g., Elasticsearch _search endpoint).
api_key	No	API key sent as Authorization: ApiKey <key>. Omit if authentication is not required.

Both features stop working and return an error if endpoint_url is not set. The api_key is optional.

---

Example: Using Both Together

A typical workflow might combine both: you use /custom-search to do a quick investigation first, then ask the agent to take action on the results — at which point it may call custom_search on its own for follow-up queries.

/custom-search service:checkout AND status:500 AND @timestamp:[now-1h TO now]

Then in a follow-up message:

Based on those results, investigate why the checkout service is failing and suggest a fix.

The agent will understand the context you provided via /custom-search and may issue additional custom_search calls autonomously as it investigates.