dspy-contextfetch

ContextFetch web search and page reading as DSPy tools.

pip install dspy-contextfetch
export CONTEXTFETCH_API_KEY="..."   # optional: without a key, requests use the keyless tier

import dspy
from dspy_contextfetch import read_page, web_search

dspy.configure(lm=dspy.LM("openai/gpt-4o-mini"))
agent = dspy.ReAct("question -> answer", tools=[web_search, read_page])
print(agent(question="What does asyncio.TaskGroup do when a task fails?").answer)

• web_search(query, num_results=5, include_domains=None, start_published_date=None): the top results as Markdown, each with title, URL, published date and the parts of the page relevant to the query.
• read_page(url, query=None): one page as Markdown; with query, only the parts relevant to it.

Both are plain functions over the contextfetch SDK, so they also work with dspy.Tool(...) or called directly. Errors from the API raise contextfetch.ContextFetchError.

Tests

uv run --group test pytest

The live test runs when ~/.contextfetch_smoke_key exists.