langchain-keenable

This package contains the LangChain integration with Keenable, a web search and page-fetch API built for AI agents.

Installation

pip install -U langchain-keenable

Optionally set a KEENABLE_API_KEY environment variable to use the authenticated endpoints. Without a key, both search and fetch transparently fall back to their keyless public endpoints.

export KEENABLE_API_KEY="your-api-key"  # optional; create one at https://keenable.ai/console

The API endpoint defaults to https://api.keenable.ai and can be overridden (e.g. for staging) with the KEENABLE_API_URL environment variable. It must be an https:// URL.

Tools

KeenableSearch

Queries the Keenable search API and returns a list of result dictionaries. All filters are per-invocation, so an agent can vary them per query:

from langchain_keenable import KeenableSearch

# Works with no key (keyless public endpoint) or with KEENABLE_API_KEY set.
tool = KeenableSearch()

results = tool.invoke({
    "query": "typescript best practices",
    "site": "github.com",          # optional: restrict to a domain
    "published_after": "2026-01-01",  # optional: YYYY-MM-DD date filters
    # "published_before" / "acquired_after" / "acquired_before" also supported
})
for result in results:
    print(result["title"], result["url"])

mode defaults to "pro" (deeper retrieval). It can be set as a class default and overridden per call.

KeenableFetch

Fetches a page via Keenable and returns its main content as markdown — pair it with KeenableSearch so an agent can read the pages it discovers:

from langchain_keenable import KeenableFetch

tool = KeenableFetch()
page = tool.invoke({"url": "https://example.com/article"})
print(page["title"], page["content"])

Error handling

Both tools set handle_tool_error = True: rate limits (429), auth (401) and credit (402) errors, network timeouts and malformed responses are surfaced to the agent as an error string (carrying the backend's message) rather than raising and crashing the agent loop.

Async

Both tools implement _arun, so await tool.ainvoke({...}) works (the request runs in a worker thread).

The tools can be bound to any LangChain chat model that supports tool calling and used within an agent.