Skip to main content
Pre-release

This release is a pre-release and may not be stable for production use.

Unique Web Search

Configurable web search tool for Unique assistants: search the web, read pages, and return citation-ready content chunks.

Outbound web access should go through Unique Search Proxy (connectors/unique_search_proxy/). When SEARCH_PROXY_BASE_URL is set, search and crawl calls use unique-search-proxy-sdk; credentials and provider egress stay on the proxy pod.

Mode When Behaviour
Proxy (recommended) SEARCH_PROXY_BASE_URL set _proxy_search / _proxy_crawl via the SDK
Legacy URL unset Direct provider calls from this process (_legacy_search / _legacy_crawl)
  • New standard engines (Brave, Perplexity) are proxy-only — legacy raises.
  • Custom API uses proxy egress only when FEATURE_FLAG_SEARCH_PROXY_ENABLE_CUSTOM_API_UN_26736 is enabled; otherwise the legacy path reports URL-safety outcomes without blocking.
  • Future provider work targets Search Proxy only; legacy is preserved for migrated engines.

Packages involved: unique-search-proxy-core (schemas), unique-search-proxy-sdk (HTTP), unique-search-proxy (server). System overview: connectors/unique_search_proxy/README.md.

Architecture

Web Search Tool Architecture

Search engine kinds

Query → normalised results (URL / title / snippet / optional content). Optional page crawl via the configured crawler.

Engine Proxy Notes
Google legacy + proxy Requires scraping by default
Brave proxy-only Rich snippets; configurable requires_scraping (on by default); ExposableParam knobs
Perplexity proxy-only Content extraction knobs; configurable requires_scraping (on by default); ExposableParam
Custom API legacy + proxy Your REST endpoint; proxy mode rejects redirects

Standard engines that live in proxy-core use ExposableParam (expose + value) for optional provider knobs. See Search Engines README.

Agent search (grounding)

An upstream AI agent / grounded model searches and returns opaque text that this tool parses into WebSearchResults. Bing results take their URLs only from the citations Foundry attaches to the answer, never from the answer text.

Engine Proxy Notes
Bing (Azure AI Foundry grounding) legacy + proxy Configurable requires_scraping; fixed market / freshness chosen by an admin, with market falling back to BING_AGENT_DEFAULT_MARKET
VertexAI (Gemini grounding) legacy + proxy Configurable requires_scraping

Bing's grounding knobs are baked into the hashed Foundry agent name, so changing one provisions a new agent version.

Web crawlers (page readers)

Crawler Auth Notes
Basic none HTTP + content-type toggles; tool-local url_blocked_patterns (legacy)
Tavily TAVILY_API_KEY Extract API
Jina JINA_API_KEY Reader API
Firecrawl FIRECRAWL_API_KEY Batch scrape

No Crawl4AI. Crawler configs have no exposable LLM knobs.

Execution modes

Mode Status Behaviour
V1 Stable Refine query → search → crawl → process
V2 Stable Parallel plan steps (search / read URL)
V3 Experimental Agent loop: snippet search vs on-demand read_urls

Proxy vs legacy is transparent to executors.

Detailed subsystem docs

Configuration

  • SEARCH_PROXY_BASE_URL — enables the Search Proxy client
  • Provider API keys (on the proxy pod when proxy is enabled; otherwise on assistants-core for legacy)
  • Search / crawler / content-processing settings via WebSearchConfig

Dependency management (uv.lock + min/latest testing)

This package is a library and uses uv for dependency management.

We run tests additionally with minimal dependencies to ensure that the listed ranges are valid. NOTE: We use lowest-direct, not lowest. Lowest attempts to use the lowest possible dependency versions transitively causing issues if a dependency has incorrect metadata. Example:

  • google-cloud-aiplatform says it works with shapely<3.0.0.
  • The lowest resolver assumes 1.0 which needs python 2 -> breaks Therefore we use lowest-direct which only sets our direct dependencies to lowest. However, this only correctly verifies our min dependencies if our code correctly lists all the required dependencies and never imports a transitive dependency. We therefore use deptry to ensure we don't use transitive dependencies and that we have no unused dependencies.

Test locally

  • Latest deps and deptry:
cd tool_packages/unique_web_search
uv sync
uv run pytest
uv run deptry
  • Min deps:
cd tool_packages/unique_web_search
uv venv
uv pip install -e . --resolution=lowest-direct
uv export --only-group dev --no-hashes | uv pip install -r -
uv run --no-sync pytest

Metadata

Release files for unique-web-search 2026.48.0.dev6

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for unique-web-search 2026.48.0.dev6
File Size Uploaded Provenance Actions
unique_web_search-2026.48.0.dev6.tar.gz 123.5 kB Without provenance Details for unique_web_search-2026.48.0.dev6.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for unique-web-search 2026.48.0.dev6
File Interpreter ABI Platform Actions
unique_web_search-2026.48.0.dev6-py3-none-any.whl Python 3 none any Details for unique_web_search-2026.48.0.dev6-py3-none-any.whl

Total release size: 308.7 kB

Release files / unique_web_search-2026.48.0.dev6.tar.gz

Download URL unique_web_search-2026.48.0.dev6.tar.gz
Size 123.5 kB
Tags Source
SHA-256 checksum
How to use checksums
bb2526bef26e45b5ef332ed82b17ede758f784d7c42e35a2af4b5f967452584b
BLAKE2b-256 checksum
How to use checksums
25754544c09084a778cb44e96c65a1ee98b3d855b0cf6c5441045b26698dc26c
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.13.7 {"installer":{"name":"uv","version":"0.13.7","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"26.04","id":"resolute","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

Release files / unique_web_search-2026.48.0.dev6-py3-none-any.whl

Download URL unique_web_search-2026.48.0.dev6-py3-none-any.whl
Size 185.1 kB
Tags Python 3
SHA-256 checksum
How to use checksums
477555b0b78e089ce17e59a4523bbad8372d92b607217cd1315c6e5c2830684b
BLAKE2b-256 checksum
How to use checksums
b25e31b96a8605fdeb0039699a92887273fa2ff4e65c5091656a76da4724e79e
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.13.7 {"installer":{"name":"uv","version":"0.13.7","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"26.04","id":"resolute","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

Security for unique-web-search 2026.48.0.dev6

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release
Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page