Deep research

Research a topic from several angles in one call. Multiple searches, deduplicated sources, the top pages fetched, and a digest with citations.

1 min read

web_search answers one query. deep_research covers a topic: it runs several searches from different angles, removes duplicate sources, fetches the top pages and returns one digest with numbered citations.

Use it

python

from shipit_agent import Agent, DeepResearchTool
from shipit_agent.llms import build_llm_from_env

research = DeepResearchTool(max_sources=8, fetch_top=4, per_page_chars=3000)
agent = Agent(llm=build_llm_from_env(), tools=[research])
print(agent.run("Compare uv and poetry for a monorepo in 2026. Cite sources.").output)

Argument	Default	What it does
`max_sources`	`8`	Most distinct sources kept
`fetch_top`	`4`	How many of them are opened and read
`per_page_chars`	`3000`	How much of each page goes into the digest
`user_agent`	`None`	The User-Agent sent when fetching

The model calls it with a topic, and optionally its own queries list to replace the default multi-angle sweep.

Use it

See also