desearch-py

Official async Python SDK for the Desearch API.

desearch-py is a thin aiohttp client for the Desearch public API. It wraps search, X, and crawl endpoints behind a single async client class and returns pydantic models for most successful responses. The package metadata currently reports version 1.2.1 and requires Python 3.9+.

Package purpose

Use this SDK when you want to call Desearch from Python without hand-rolling HTTP requests, auth headers, or response parsing. The repository ships:

• an async client in desearch_py/api.py
• typed models and enums in desearch_py/models.py
• lightweight Sphinx scaffolding in docs/
• packaging metadata in both pyproject.toml and setup.py

Two names matter:

• package on PyPI and in Poetry metadata: desearch-py
• import path in Python code: desearch_py

Quick Start

Install

pip install desearch-py

Run a basic AI search

import asyncio
from desearch_py import Desearch


async def main() -> None:
    async with Desearch(api_key="your_api_key") as client:
        result = await client.ai_search(
            prompt="What happened in Bittensor this week?",
            tools=["web", "twitter", "reddit"],
            count=10,
        )
        print(result)


asyncio.run(main())

Read optional response cost metadata

Default calls continue to return the original typed payload only:

result = await client.ai_search(
    prompt="What happened in Bittensor this week?",
    tools=["web", "twitter"],
)
print(result.text)

When you want per-request cost visibility, opt in with include_metadata=True:

response = await client.ai_search(
    prompt="What happened in Bittensor this week?",
    tools=["web", "twitter"],
    include_metadata=True,
)

print(response.data.text)
print(response.metadata.cost_usd)
print(response.metadata.usage_count)
print(response.metadata.service)
print(response.metadata.currency)

The SDK reads metadata from the same API response headers (X-Desearch-Cost-Usd, for example 0.00015, plus X-Desearch-Usage-Count, X-Desearch-Service, and X-Desearch-Currency). Missing or malformed numeric headers are exposed as None instead of breaking a successful API call.

Reuse the client manually

import asyncio
from desearch_py import Desearch


async def main() -> None:
    client = Desearch(api_key="your_api_key")
    try:
        tweets = await client.x_search(query="desearch", sort="Latest", count=5)
        print(tweets)
    finally:
        await client.close()


asyncio.run(main())

Requirements

Dependencies declared in source:

• Python >=3.9 (pyproject.toml, setup.py)
• aiohttp >=3.8
• pydantic >=2.0

setup.py also lists typing-extensions as an install dependency.

Commands

A minimal unittest suite covers response metadata behavior. There is still no dedicated lint or format command configured in the repo.

Tech stack

• Python 3.9+
• aiohttp for async HTTP
• pydantic v2 models for typed responses
• Poetry + setuptools metadata side by side
• Sphinx scaffolding for docs

Supported SDK surface

AI search

• ai_search
• ai_web_links_search
• ai_x_links_search

X / Twitter endpoints

• x_search
• x_posts_by_urls
• x_post_by_id
• x_posts_by_user
• x_post_retweeters
• x_user_posts
• x_user_replies
• x_post_replies
• x_trends

Web endpoints

• web_search
• web_crawl

Typed exports

The package re-exports the async client plus enums and models from desearch_py.models, including Tool, WebTool, DateFilter, ResultType, Sort, ResponseData, DesearchCostMetadata, DesearchResponse, TwitterScraperTweet, WebSearchResponse, WebSearchResultsResponse, XLinksSearchResponse, XRetweetersResponse, XUserPostsResponse, and XTrendsResponse.

Architecture overview

The SDK is intentionally small.

• desearch_py/api.py contains the Desearch client and every HTTP method.
• desearch_py/models.py contains enums and permissive pydantic schemas.
• desearch_py/__init__.py re-exports the public API.

Key design decisions visible in code:

• the HTTP session is created lazily on first use
• auth is sent as Authorization: <api_key> with no Bearer prefix
• most requests share a single helper with a fixed 120 second timeout
• successful responses can opt in to a DesearchResponse wrapper with parsed cost metadata from response headers
• x_posts_by_urls and web_crawl bypass that shared helper and make direct requests, but still support the metadata wrapper
• some endpoints fall back to raw dict values instead of failing model parsing
• ai_search always sends "streaming": False

Usage notes

Use async with when possible

async with Desearch(api_key="your_api_key") as client:
    results = await client.web_search(query="Desearch API")

This is the safest pattern because it guarantees the underlying aiohttp.ClientSession gets closed.

Expect mixed return types on some methods

The following methods may return a typed object on success or a raw dict when the API shape does not match the local parser exactly:

• ai_search
• x_search
• x_posts_by_user
• x_user_replies
• x_post_replies

If your app depends on strict typing, add runtime checks before dereferencing attributes.

Override the base URL for non-production environments

client = Desearch(
    api_key="your_api_key",
    base_url="https://staging-api.desearch.ai",
)

Repo boundaries

Always:

• treat this repo as the Python SDK only
• verify docs against desearch_py/api.py, desearch_py/models.py, pyproject.toml, and setup.py
• document current behavior, even when it exposes limitations

Never:

• describe unsupported streaming or retry behavior as implemented
• assume generated files in docs/_build/ are the source of truth
• confuse this repo with desearch.js, desearch-public-api, or desearch-console

Additional docs

• Feature inventory
• Architecture notes
• Known issues
• ADR: current behavior is the documentation source of truth

Links

• Desearch: https://desearch.ai
• Console: https://console.desearch.ai
• API docs: https://desearch.ai/docs/api-reference/