POST /v1/responses
Web search is a tool entry in the tools array. Per-request allow lists use OpenAI’s native tools_options.web_search.filters.
Request
from openai import OpenAI
client = OpenAI(base_url="https://api.abliteration.ai/v1", api_key=os.environ["ABLIT_KEY"])
resp = client.responses.create(
model="abliterated-model",
input="Latest SEC filings for NVDA",
tools=[{"type": "web_search"}],
)
Accepted type aliases: web_search, web_search_preview, web_search_2025_08_26, web_search_preview_2025_03_11.
With an allow list (OpenAI native shape)
resp = client.responses.create(
model="abliterated-model",
input="Latest SEC filings for NVDA",
tools=[{"type": "web_search"}],
extra_body={
"tools_options": {
"web_search": {
"filters": {
"include_domains": ["sec.gov", "nvidia.com"],
}
}
}
},
)
Up to 100 domains. Omit http:// / https:// prefixes.
| Filter | OpenAI native | Honored by abliteration.ai |
|---|
tools_options.web_search.filters.include_domains | Yes | Yes |
Block list (exclude_domains) | Not in OpenAI spec | Only via Policy Gateway project-level |
Forced search and lifecycle events
When combined with tool_choice={"type": "required"}, the search fires before the model starts. The stream emits five lifecycle events up front:
response.output_item.added (web_search_call, status: in_progress)
response.web_search_call.in_progress
response.web_search_call.searching
response.web_search_call.completed
response.output_item.done
Then the model stream begins.
Project-level enforcement
Project-level allow/block lists configured via Policy Gateway cap the request:
- Allow: request
include_domains ⊆ project allowed_domains. A request asking for a domain outside the project list returns 400.
- Block: project
blocked_domains always applied (OpenAI has no per-request block shape; the project list is the only way to exclude domains on this surface).
If the project has neither list configured, per-request filters pass through unchanged.Last modified on May 3, 2026