crewai-agentsearch
CrewAI tools for AgentSearch: live web search for agents. Free search needs no API key.
| Tool | Name | Cost |
|---|---|---|
AgentSearchTool |
agentsearch_search |
Free (up to 25 searches/day per IP, up to 5 results, no key). Paid and unlimited only when you opt in (see Paid mode). |
AgentSearchNewsTool |
agentsearch_news_search |
$0.003/call (recent news with publish dates; no results, no charge) |
AgentSearchExtractTool |
agentsearch_extract |
$0.002/call (USDC on Base via x402 to AgentSearch) |
AgentSearchRenderTool |
agentsearch_render |
$0.005/call (headless browser for JS pages) |
AgentSearchScreenshotTool |
agentsearch_screenshot |
$0.004/call (PNG/JPEG saved to a local file; path returned) |
AgentSearchSearchAndReadTool |
agentsearch_search_and_read |
$0.005/call (search plus the top pages' text in one payment) |
Install
pip install -U crewai-agentsearch
Quickstart
from crewai import Agent, Crew, Task
from crewai_agentsearch import AgentSearchTool
search = AgentSearchTool()
print(search.run(query="latest CrewAI release", max_results=3)) # JSON string
researcher = Agent(role="Researcher", goal="Find current facts with sources",
backstory="You always cite URLs.", tools=[search])
task = Task(description="What changed in the latest CrewAI release?", expected_output="3 bullets with URLs", agent=researcher)
print(Crew(agents=[researcher], tasks=[task]).kickoff())
Add AgentSearchNewsTool(), AgentSearchExtractTool(), AgentSearchRenderTool(), AgentSearchScreenshotTool() and AgentSearchSearchAndReadTool() to tools once paid mode is set up.
Filters, news and screenshots (0.5.0)
from crewai_agentsearch import AgentSearchNewsTool, AgentSearchScreenshotTool, AgentSearchTool
AgentSearchTool(mode="paid").run(query="x402 payments", freshness="week", include_domains=["github.com"])
AgentSearchNewsTool().run(query="Base network USDC", freshness="day", country="us", max_results=8)
AgentSearchScreenshotTool().run(url="https://example.com", viewport={"width": 390, "height": 844})
# -> JSON with screenshot.path (image file) instead of base64, so the image never floods the model context
Paid search, news and search-and-read accept freshness (day/week/month/year), date_from/date_to (YYYY-MM-DD), country, language, and up to 5 include_domains/exclude_domains. Free search ignores filters. Extract accepts max_chars (up to 50000) and include_links. Screenshots are saved in AGENTSEARCH_SCREENSHOT_DIR (default: a temp folder).
Paid mode
Install the paid extra (pip install "crewai-agentsearch[paid]") and set AGENTSEARCH_WALLET_PRIVATE_KEY (or POCKET_PRIVATE_KEY) to the private key of a new wallet made only for this, funded with a little USDC on Base (e.g. $1; no ETH needed). Since 0.3.0 each call is paid over x402 directly to AgentSearch's own API (https://api.agentsearchhq.com, USDC on Base, settled only on success) using the official x402 Python SDK: search $0.003, news $0.003, extract $0.002, render $0.005, screenshot $0.004, search-and-read $0.005. $1 covers about 500 extracts or 200 search-and-read calls. Prefer the Pocket Network Agentic Portal ($0.005 per call)? Set AGENTSEARCH_PAYMENT_ROUTE=pocket.
News, extract, render, screenshot and search-and-read are always paid and need the key. News, screenshot and search-and-read exist only on the direct route (not on Pocket).
Paid search is opt-in (changed in 0.2.1). Setting a wallet key does not make search paid. AgentSearchTool (default mode="auto") stays on free search until you opt in in one of two ways:
AgentSearchTool(mode="paid") # always paid search
export AGENTSEARCH_MODE=paid # opt in for AgentSearchTool in the default mode="auto"
| Setup | AgentSearchTool search |
|---|---|
mode="auto" (default), no AGENTSEARCH_MODE |
Free, even if a wallet key is set |
mode="auto" + AGENTSEARCH_MODE=paid |
Paid |
mode="auto" + paid opted in or a key set, but [paid] extra not installed |
Free, with a logged warning |
mode="paid" |
Paid; without the [paid] extra it returns an error that tells you how to install it |
mode="free" |
Always free |
| Env var | Default |
|---|---|
AGENTSEARCH_MODE |
unset (paid opts mode="auto" in to paid search; free forces free) |
AGENTSEARCH_MAX_TOTAL_ATOMIC |
1000000 ($1.00 per process) |
AGENTSEARCH_PAYMENT_ROUTE |
direct (AgentSearch's x402 API); pocket = Pocket Network Agentic Portal at $0.005/call |
AGENTSEARCH_MAX_PER_CALL_ATOMIC |
20000 on the direct route, 5000 on Pocket ($0.020 / $0.005; can only be lowered) |
AGENTSEARCH_SCREENSHOT_DIR |
a temp folder (agentsearch-screenshots) |
AGENTSEARCH_TIMEOUT_MS |
30000 |
AGENTSEARCH_EXTRACT_RENDER_FALLBACK |
unset (1 turns on render fallback for extract, see below) |
Limits are checked before anything is signed. The key is never returned in tool output. Paid results include a payment object.
Extract to render fallback
Set AGENTSEARCH_EXTRACT_RENDER_FALLBACK=1 to make AgentSearchExtractTool automatically retry a URL with render when extract returns thin content (under 200 characters of text, typical for JavaScript-heavy pages). It is off by default because the render retry is a second paid call ($0.005, so up to $0.007 for that URL), still bounded by AGENTSEARCH_MAX_TOTAL_ATOMIC. You can also turn it on or off per call with the tool's render_fallback argument, which overrides the env var. When the fallback runs, the result includes a render_fallback object.
Errors
Tools return "Error: ..." strings instead of raising. A 429 from the free relay means the free daily limit (up to 25 searches/day per IP) was hit, and the message explains how to upgrade. Search, news and search-and-read reject queries shorter than 2 characters (after trimming) before any request, so nothing is sent or charged; other 4xx errors from the free relay say to fix the input instead of retrying it unchanged.
See CHANGELOG.md for release notes.
Development
pip install -e ".[test]" && pytest # unit tests (mocked)
AGENTSEARCH_LIVE=1 pytest -k live # one real free search
Docs: https://agentsearchhq.com/agents.md?utm_source=crewai-agentsearch&utm_medium=package
MIT
Metadata
Release files for crewai-agentsearch 0.5.3
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| crewai_agentsearch-0.5.3.tar.gz | 21.3 kB | Without provenance | Details for crewai_agentsearch-0.5.3.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| crewai_agentsearch-0.5.3-py3-none-any.whl | Python 3 | none | any | Details for crewai_agentsearch-0.5.3-py3-none-any.whl |
Total release size: 39.0 kB
Release files / crewai_agentsearch-0.5.3.tar.gz
| Download URL | crewai_agentsearch-0.5.3.tar.gz |
|---|---|
| Size | 21.3 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
2dfecc03e008abcd976b047b09fd437c4fbbc67d52559990446a17351e7d9b71
|
|
BLAKE2b-256 checksum How to use checksums |
21298acf5de2e43c74f719bb102ca82726f921846bb1326481fe1dd5d37f0fdd
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.5
|
Release files / crewai_agentsearch-0.5.3-py3-none-any.whl
| Download URL | crewai_agentsearch-0.5.3-py3-none-any.whl |
|---|---|
| Size | 17.7 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ee6cbf6f57a28020f0da185cd4a3e63519687c4f44f46a73cbadbf3badb47df6
|
|
BLAKE2b-256 checksum How to use checksums |
8633af0f173f4825595b426ef123c00856a8c8f28e1edefad730f3377ed9778d
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.13.5
|
Security for crewai-agentsearch 0.5.3
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.