> ## Documentation Index
> Fetch the complete documentation index at: https://docs.abliteration.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Web search on abliteration.ai

> Enable web search on abliteration.ai across OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages — citations included.

**To enable web search on abliteration.ai**, turn it on under Account → Web tools in the [console](https://abliteration.ai/console), then send requests using the native web-search shape for whichever API surface you're calling. The model returns inline citations for any web sources it draws from.

<CardGroup cols={3}>
  <Card title="OpenAI Chat Completions" icon="globe" href="/capabilities/web-search/openai-chat-completions">
    Enable search directly on the request body.
  </Card>

  <Card title="OpenAI Responses" icon="globe" href="/capabilities/web-search/openai-responses">
    Add search as a tool, with optional allow-list filters.
  </Card>

  <Card title="Anthropic Messages" icon="globe" href="/capabilities/web-search/anthropic-messages">
    Add search as a tool, with allow and block lists plus location.
  </Card>
</CardGroup>

## Enable web search

### Any account

Turn on web search under **Account → Web tools** in the [console](https://abliteration.ai/console). Once on, pick the surface above and send requests using its native web-search shape. Per-request domain filters (where the surface supports them) pass through unchanged.

### Policy Gateway accounts

Everything above, plus per-project domain ceilings. Under **Project → Web tools**, set `allowed_domains` and/or `blocked_domains` on a project and they apply to every request made with that project's key.

These lists live on the project (not the policy).

#### How ceilings merge with per-request filters

When both a project ceiling and a per-request filter are set:

* **Allow lists** are intersected. The request's allow list must be a subset of the project's. Asking for a domain outside the project list returns `400`.
* **Block lists** are unioned. Project blocks always apply; request blocks (where the surface supports them) apply on top.

What each surface sends natively:

* **OpenAI Chat Completions** has no per-request allow or block field — the project ceiling is the only domain control.
* **OpenAI Responses** sends `include_domains` (allow) only — block comes from the project.
* **Anthropic Messages** sends both `allowed_domains` and `blocked_domains`.

If the project has neither list configured, per-request filters pass through unchanged.

## In practice

Web search works inside any client that talks to abliteration.ai.

### Codex CLI

<Frame>
  <img src="https://mintcdn.com/acme-c0f07daa/esggkRbkZyEOfVDk/images/codex-web-search.png?fit=max&auto=format&n=esggkRbkZyEOfVDk&q=85&s=42962ab7b7a33ac15ed16945defbb3e1" alt="OpenAI Codex CLI calling web_search via abliteration.ai with a cited URL in the response" width="1570" height="1005" data-path="images/codex-web-search.png" />
</Frame>

Enable by adding `web_search = "live"` to the abliteration profile in `~/.codex/config.toml`. Full setup → [Codex integration](/integrations/codex).

### Claude Code

<Frame>
  <img src="https://mintcdn.com/acme-c0f07daa/esggkRbkZyEOfVDk/images/claude-code-web-search.png?fit=max&auto=format&n=esggkRbkZyEOfVDk&q=85&s=4f9e7c0c0f60f7480616ba23722a0450" alt="Anthropic Claude Code calling web_search via abliteration.ai with citations" width="1658" height="1019" data-path="images/claude-code-web-search.png" />
</Frame>

Full setup → [Claude Code integration](/integrations/claude-code).