Add KeenableSearchTool (keyless web search)

[ilya-bogin-keenable]

Adds a KeenableSearchTool backed by the Keenable Search API.

Unlike the key-gated backends (GoogleSearchTool, ApiWebSearchTool) it is keyless by default — the public endpoint needs no account or API key (rate-limited to 1,000 req/hour). An optional KEENABLE_API_KEY only lifts the rate limit; it is never required. It uses the shared web_search tool name and mirrors ApiWebSearchTool's shape (per-second rate limit + markdown output), so agents select it like any other search backend:

from smolagents import CodeAgent, KeenableSearchTool, InferenceClientModel

agent = CodeAgent(tools=[KeenableSearchTool()], model=InferenceClientModel())  # keyless
agent.run("What are the latest developments in retrieval-augmented generation?")

Context: the maintainer of txtai suggested smolagents as the right home for this (neuml/txtai#1127) — txtai already adapts smolagents tools, so this unblocks that ecosystem too.

Changes:

• default_tools.py — KeenableSearchTool + __all__ export (not added to TOOL_MAPPING since it shares the web_search name, same as ApiWebSearchTool/GoogleSearchTool)
• docs/source/en/reference/default_tools.md — reference entry
• tests/test_default_tools.py — keyless vs keyed endpoint selection + no-results handling

Verified: 3 new tests pass, ruff clean, and the keyless endpoint was self-tested live.

🤖 Generated with Claude Code