Skip to main content
Pre-release

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

unique-search-proxy-sdk

Part of Unique Search Proxy · PyPI: unique-search-proxy-sdk


1. What this package is

The SDK is the caller's HTTP client. Platform services (e.g. assistants-core) use it to reach the proxy pod at runtime — search, agent search, and crawl — without hand-crafting httpx calls or duplicating route paths.

It depends on core for request validation and typed errors. It does not include the FastAPI server.

Package Question it answers
Core What can be configured and what does a valid request/response look like?
Client How are provider calls executed at runtime?
SDK (this) How do callers reach the proxy over HTTP?

2. Role in the system

The SDK implements Path B (runtime HTTP) from the system overview. Schema work (Path A) stays in core — the SDK does not expose deployment config or LLM call-schema endpoints.

flowchart LR
    Caller["Caller service"]
    SDK["unique_search_proxy_sdk"]
    Core["unique_search_proxy_core"]
    Proxy["unique_search_proxy_client"]

    Caller --> SDK
    SDK --> Core
    SDK -->|"httpx"| Proxy

Before a search call, the caller (or SDK) validates the payload against core request models. Non-2xx responses become the same ProxyError subclasses defined in core.


3. Architecture

Two layers: a hand-written facade over OpenAPI-generated code.

flowchart TB
    subgraph sdk_pkg["unique_search_proxy_sdk"]
        Facade["UniqueSearchProxyClient"]
        SC["SearchClient"]
        ASC["AgentSearchClient"]
        CC["CrawlClient"]
        Transport["OpenapiTransport"]
        Gen["_generated/"]
        Conv["converters.py"]
        Err["errors.py"]
    end

    Core["unique_search_proxy_core"]
    Proxy["Proxy pod"]

    Facade --> SC & ASC & CC
    SC & ASC & CC --> Transport
    Transport --> Gen
    SC --> Conv --> Core
    Err --> Core
    Gen -->|"httpx AsyncClient"| Proxy
Layer Path Responsibility
Facade client.py UniqueSearchProxyClient — composes sub-clients, health(), ready()
Sub-clients search_client.py, agent_search_client.py, crawl_client.py Per-provider typed endpoints + compatibility dispatchers
Endpoint factory _endpoint.py async_post_endpoint / async_sse_endpoint (ParamSpec from core request models)
Transport _transport.py, _http.py Shared httpx lifecycle, base URL, timeout
Generated _generated/ Route functions + attrs models from OpenAPI (do not edit)
Converters converters.py Core Pydantic models → generated SDK models
Errors errors.py Error envelope → core ProxyError subclasses

The client package is the source of truth for the HTTP contract. When routes change, regenerate _generated/ from unique_search_proxy_client/openapi.json.


4. Usage

Each provider exposes a typed endpoint whose kwargs mirror the core *Request Pydantic model (via ParamSpec). Snake_case and camelCase aliases both work (fetch_size / fetchSize).

from unique_search_proxy_sdk import UniqueSearchProxyClient

async with UniqueSearchProxyClient("http://unique-search-proxy:2349") as client:
    await client.health()

    # Typed per-provider endpoints (preferred)
    result = await client.search.google(query="unique ag", gl="ch", fetch_size=10)
    brave = await client.search.brave(query="news", country="US", safesearch="strict")
    agent = await client.agent_search.bing(query="EU AI Act timeline", fetch_size=5)
    crawl = await client.crawl.basic(urls=["https://example.com"])

    # Compatibility dispatchers (engine/crawler string)
    result = await client.search.search("unique ag", engine="google", fetch_size=10)
    agent = await client.agent_search.search("query", engine="bing")
    crawl = await client.crawl.crawl(["https://example.com"], crawler="Basic")

    # Agent SSE streaming
    async for event in client.agent_search.bing_stream(query="query"):
        ...

    # Low-level: one generated function per route
    raw_client = client.openapi

Search payloads are validated through core's parse_search_request() before serialization. Agent and crawl clients follow the same pattern.


5. API mapping

Facade method HTTP
health() GET /health
ready() GET /ready
search.google(...) / .brave(...) / .perplexity(...) POST /v1/search
search.search(...) POST /v1/search (dispatcher)
agent_search.bing(...) / .vertexai(...) POST /v1/agent-search
agent_search.bing_stream(...) / .vertexai_stream(...) POST /v1/agent-search/stream (SSE)
agent_search.search(...) / .stream(...) same routes (dispatchers)
crawl.basic(...) / .tavily(...) / .jina(...) / .firecrawl(...) POST /v1/crawl
crawl.crawl(...) POST /v1/crawl (dispatcher)
openapi Low-level generated client (one function per route)

Endpoint payloads and provider ids → Client README.


6. Error handling

Non-success responses with a proxy error envelope raise typed exceptions from core:

from unique_search_proxy_core import EngineNotConfiguredError, UpstreamError

ENGINE_NOT_CONFIGURED (503) includes missing_env_vars parsed from the error details list — useful for operators and LLM tool consumers.

Transport failures (non-JSON body, connection errors) raise UniqueSearchProxyClientError.


7. Codegen workflow

Generated via openapi-python-client from the server's OpenAPI spec.

cd ../unique_search_proxy_client
uv sync
uv run python scripts/generate_sdk.py
Path Role
unique_search_proxy_sdk/_generated/ Regenerated httpx client + attrs models
unique_search_proxy_client/openapi.json Exported spec (codegen input)
unique_search_proxy_sdk/client.py Hand-written facade (not regenerated)

Other codegen tools considered: OpenAPI Generator, datamodel-code-generator, Kiota — see git history in the client README if needed.


8. Testing

Pass an injected httpx.AsyncClient with ASGITransport for in-process tests against create_app():

import httpx
from httpx import ASGITransport
from unique_search_proxy_client.web.app import create_app
from unique_search_proxy_sdk import UniqueSearchProxyClient

app = create_app()
async with httpx.AsyncClient(
    transport=ASGITransport(app=app),
    base_url="http://test",
) as http:
    async with UniqueSearchProxyClient("http://test", http_client=http) as client:
        ...

Run the app lifespan so the in-process HttpClientPool is initialised.


9. Installation & development

cd unique_search_proxy_sdk
uv sync
uv run pytest
uv run ruff check .
uv run basedpyright

Depends on unique-search-proxy-core. Does not depend on unique-search-proxy at runtime (only as a dev dependency for integration tests).


License

Proprietary — Unique AG

Metadata

Release files for unique-search-proxy-sdk 2026.44.0.dev4

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-search-proxy-sdk 2026.44.0.dev4
File Size Uploaded Provenance
unique_search_proxy_sdk-2026.44.0.dev4.tar.gz 41.4 kB Without provenance Details

Built distribution (wheel)

Table of built distributions (wheels) for unique-search-proxy-sdk 2026.44.0.dev4
File Interpreter ABI Platform
unique_search_proxy_sdk-2026.44.0.dev4-py3-none-any.whl Python 3 none any Details

Total release size: 125.0 kB

Release files / unique_search_proxy_sdk-2026.44.0.dev4.tar.gz

Download URL unique_search_proxy_sdk-2026.44.0.dev4.tar.gz
Size 41.4 kB
Tags Source
SHA-256 checksum
How to use checksums
e163405607c8c30edd840a95e177fed30bbdcf0a1c974ac0d95d6aad3078f28e
BLAKE2b-256 checksum
How to use checksums
28b57c8d72efd66c48d9e753edabe9b2ce56e6b6ac7082a6f37714d6d61c6304
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

Release files / unique_search_proxy_sdk-2026.44.0.dev4-py3-none-any.whl

Download URL unique_search_proxy_sdk-2026.44.0.dev4-py3-none-any.whl
Size 83.6 kB
Tags Python 3
SHA-256 checksum
How to use checksums
3aad472bfcb58bd8a8cc16b442138ab073b4c5781fade239db23daae3f648403
BLAKE2b-256 checksum
How to use checksums
cb26d384ad145a8575a5c1783c903ef39d74ec25f587b720cf67f35f9018244b
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.11.28 {"installer":{"name":"uv","version":"0.11.28","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}

Security for unique-search-proxy-sdk 2026.44.0.dev4

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.

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