agentsearch-client

agentsearch

Python client for the AgentSearch API — self-hosted, multi-engine web search for AI agents and RAG pipelines.

Zero dependencies required. Uses urllib out of the box; install httpx for connection pooling.

Install

pip install agentsearch

# Optional: better HTTP performance
pip install agentsearch[httpx]

Quick Start

from agentsearch import AgentSearch

client = AgentSearch("http://localhost:3939")

# Search the web
results = client.search("python async patterns", count=5)
for r in results.results:
    print(f"{r.title}: {r.url}")

API Reference

AgentSearch(base_url, timeout=30.0)

Create a client instance. If httpx is installed, it uses connection pooling automatically.

client = AgentSearch("http://localhost:3939", timeout=15.0)

Supports context manager:

with AgentSearch("http://localhost:3939") as client:
    results = client.search("query")

---

client.search(query, *, count=10, engines=None, domain=None, exclude_domains=None, fetch=False)

Multi-engine web search with deduplication and scoring.

# Basic search
results = client.search("python async patterns", count=5)

# Search + extract page content
results = client.search("python async", count=5, fetch=True)
for r in results.results:
    print(r.content[:200] if r.content else "No content")

# Filter to specific domain
results = client.search("site reliability", domain="google.com")

---

client.search_extract(query, *, count=5, engines=None)

Search and automatically extract content from top results via the kill chain.

results = client.search_extract("RAG pipeline best practices", count=3)
for r in results.results:
    print(f"{r.title}\n{r.content[:500]}\n")

---

client.deep_search(query, *, count=10)

Multi-query fusion search — generates 3-5 query variations, merges and deduplicates.

results = client.deep_search("manufacturing OEE best practices", count=10)
print(f"Queries used: {results.meta.queries_used}")

---

client.search_policy(query, *, count=10, fetch=False)

Policy-optimized search with think tank boosting and junk filtering.

results = client.search_policy("South China Sea maritime disputes", fetch=True)

---

client.read(url, *, max_chars=None, skip_cache=False)

Extract readable content from any URL using 9 escalating strategies (direct → readability → UA rotation → Wayback Machine → Google Cache → search-about → custom adapters → PDF → YouTube).

content = client.read("https://example.com/article")
print(content.content)
print(f"Strategy: {content.strategy}, Chars: {content.chars}")

---

client.read_batch(urls, *, max_chars=None)

Extract content from multiple URLs concurrently (max 20).

results = client.read_batch([
    "https://example.com/page1",
    "https://example.com/page2",
])
print(f"Success: {results.successful}/{results.total}")

---

client.news(query, *, count=10, engines=None)

Search news across 9+ engines (Google News, Bing News, Reuters, Yahoo, Brave, etc.).

articles = client.news("AI agents", count=10)
for a in articles.results:
    print(f"[{a.source}] {a.title} ({a.published})")

---

client.search_jobs(query, *, location=None, salary_min=None)

Search jobs across LinkedIn, Indeed, Glassdoor, and ZipRecruiter.

jobs = client.search_jobs("python engineer", location="Chicago", salary_min=120000)
for j in jobs.results:
    sal = f"${j.salary_min:,}-${j.salary_max:,}" if j.salary_min else "Not listed"
    print(f"{j.title} @ {j.source} — {sal}")

---

client.health()

Check API health and SearXNG connectivity.

h = client.health()
print(f"Status: {h.status}, Version: {h.version}")

---

Module-level convenience functions

For quick scripts, use module-level functions with a default client:

import agentsearch

agentsearch.configure("http://localhost:3939")

results = agentsearch.search("python")
content = agentsearch.read("https://example.com")
articles = agentsearch.news("AI")
jobs = agentsearch.search_jobs("data engineer")

Error Handling

from agentsearch import AgentSearch, AgentSearchError

client = AgentSearch("http://localhost:3939")

try:
    results = client.search("query")
except AgentSearchError as e:
    print(f"Error: {e}")
    print(f"Status code: {e.status_code}")
    print(f"Detail: {e.detail}")

Why AgentSearch?

Self-Hosted Server

This package is just the client. To run your own AgentSearch server:

git clone https://github.com/brcrusoe72/agent-search.git
cd agent-search
docker compose up -d
# API available at http://localhost:3939

See the server repository for full setup instructions.

License

MIT