examples

langchain-diffbot examples

Two ways to see the package in action. Both expect DIFFBOT_API_TOKEN and (for the agent code) ANTHROPIC_API_KEY in the environment — copy .env.example to .env and fill in the values, or export them in your shell.

Install

From inside this repo (the common case — [tool.uv.sources] already points diffbot-python at its GitHub source):

uv sync --extra examples

From outside the repo (e.g. after langchain-diffbot is published to PyPI) you'd install both explicitly since diffbot-python isn't on PyPI yet:

pip install \
    "diffbot-python @ git+https://github.com/diffbot/diffbot-python" \
    "langchain-diffbot[examples]"

Notebook

quickstart.ipynb is a full tour of the package:

1. Knowledge Graph retriever + output shaping
2. Native async
3. Web Search retriever
4. Extract tool + extract loader
5. Entities tool
6. ChatDiffbot (Diffbot's own LLM with native streaming)
7. Bring-your-own SDK client
8. A multi-tool research agent that uses KG search + web search + URL extract

# uv-managed (recommended — handles PATH automatically):
uv run --with jupyter jupyter lab examples/quickstart.ipynb

# Or plain pip (use `python -m jupyter`, not `jupyter`, to avoid PATH issues):
pip install jupyter
python -m jupyter lab examples/quickstart.ipynb

The notebook is regenerated from _build_notebook.py — edit that file (cell sources are inline) and re-run uv run python examples/_build_notebook.py rather than editing the .ipynb directly.

CLI

company_research/ is the same multi-tool agent packaged as a one-shot CLI. Useful for shell scripting or quick spot checks.

cd examples
python -m company_research "What companies in Austin work on robotics?"
python -m company_research --quiet "Who are the executives at Diffbot?"
python -m company_research "What did Diffbot announce most recently?"

The agent has three tools:

• search_kg(dql_query) — Knowledge Graph search
• web_search(query) — natural-language web search
• extract_url(url) — fetch and read a single page

It picks its own approach, may iterate, and cites the entity IDs / URLs it used. Drop --quiet to see the tool calls and intermediate responses.

Model

Defaults to anthropic:claude-haiku-4-5 because a multi-step agent loop on a fresh Anthropic account can blow past Sonnet's 30k input-tokens-per-minute Tier 1 limit. Override with:

COMPANY_RESEARCH_MODEL=anthropic:claude-sonnet-4-6 python -m company_research "..."

Web app

dql_explorer/ is a browser UI for the DQL-authoring loop: type a question in plain English, and an agent inspects the ontology, probes query variants, writes the DQL, and the results come back as a table. It's a FastAPI backend serving a React + TypeScript (Vite) frontend, with optional LangSmith tracing. See dql_explorer/README.md for setup — in short:

cd dql_explorer/web && pnpm install && pnpm build && cd ..
uv run --extra examples python -m dql_explorer   # then open http://127.0.0.1:8000