livetennisapi-haystack

Haystack 2.x integration for the Live Tennis API: live scores, matches and players as Haystack Documents for RAG and agent pipelines.

• LiveTennisMatchFetcher — live / upcoming / completed matches (optionally one match by id, optionally filtered by tour) as Documents. content is a clean human-readable match summary; meta carries the structured fields (ids, players, sets/games/points, server, winner).
• LiveTennisPlayerSearch — player search by name, ranked players first, same Document shape.

Built on the official livetennisapi Python client (retries, error mapping, typed models) — no hand-rolled HTTP.

Installation

pip install livetennisapi-haystack

You need a Live Tennis API key (free tier: 1000 requests/day, 30/min). Export it as an environment variable — the components read LIVETENNISAPI_KEY by default and never accept a plain-string key:

export LIVETENNISAPI_KEY="your-key"

Usage

Standalone

from livetennisapi_haystack import LiveTennisMatchFetcher

fetcher = LiveTennisMatchFetcher()          # key from LIVETENNISAPI_KEY
result = fetcher.run(status="live", limit=5)
for doc in result["documents"]:
    print(doc.content)
    # e.g. "Carlos Alcaraz (ESP, #2) vs Jannik Sinner (ITA, #1) — match at Wimbledon,
    #       grass court, round QF, best of 5. Live now. Score: sets 1-1, games 6-4, 3-6,
    #       2-1, points 30-15. Carlos Alcaraz (ESP, #2) is serving."

In a pipeline (runnable with only LIVETENNISAPI_KEY)

from haystack import Pipeline

from livetennisapi_haystack import LiveTennisMatchFetcher, LiveTennisPlayerSearch

pipe = Pipeline()
pipe.add_component("matches", LiveTennisMatchFetcher(limit=5))
pipe.add_component("players", LiveTennisPlayerSearch(limit=3))

result = pipe.run({"matches": {"status": "live"}, "players": {"query": "alcaraz"}})
for doc in result["matches"]["documents"] + result["players"]["documents"]:
    print("-", doc.content)

RAG over live scores

from haystack import Pipeline
from haystack.components.builders.chat_prompt_builder import ChatPromptBuilder
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.dataclasses import ChatMessage

from livetennisapi_haystack import LiveTennisMatchFetcher

prompt_template = [
    ChatMessage.from_system("You are a tennis commentator."),
    ChatMessage.from_user(
        "Current matches:\n"
        "{% for document in documents %}{{ document.content }}\n{% endfor %}\n"
        "Answer the following question: {{ query }}\nAnswer:"
    ),
]

pipe = Pipeline()
pipe.add_component("matches", LiveTennisMatchFetcher(limit=10))
pipe.add_component("prompt_builder", ChatPromptBuilder(template=prompt_template, required_variables={"query", "documents"}))
pipe.add_component("llm", OpenAIChatGenerator(model="gpt-4o-mini"))
pipe.connect("matches.documents", "prompt_builder.documents")
pipe.connect("prompt_builder.prompt", "llm.messages")

query = "Who is closest to winning right now?"
result = pipe.run({"matches": {"status": "live"}, "prompt_builder": {"query": query}})
print(result["llm"]["replies"][0].text)

A complete runnable script lives at examples/live_demo.py.

Behavior worth knowing

• 403 tier wall: when your key is valid but the plan does not unlock an endpoint, the component returns a single readable Document (tagged meta["error"] = "upgrade_required") instead of raising — an agent can tell the user; a RAG pipeline can filter it out. All other errors (bad key, network down, rate limit) still raise the official client's typed exceptions.
• Sparse data is normal: score.server is nullable (between points the feed may not know who serves next — the summary simply omits the serving sentence), doubles teams have no individual rankings and a null data_completeness, and points are strings ("0", "15", "30", "40", "AD"). The components tolerate all of it.
• Serialization: both components implement to_dict/from_dict; the API key is stored as a Secret environment-variable reference, never as a value, so pipelines serialize safely to YAML. Note that Haystack 3.0 refuses to deserialize third-party components unless their module is allow-listed, so reload pipelines with Pipeline.loads(yaml_str, allowed_modules=["livetennisapi_haystack.match_fetcher", "livetennisapi_haystack.player_search"]) (or haystack.core.serialization.allow_deserialization_module(...)).
• tour filter: the API's documented tour query parameter is not yet exposed by livetennisapi 1.0.2's list_matches(), so the component routes that one call through the official client's transport layer (same auth/retries/error mapping).
• Sync only for now: run() — no run_async yet, although the official client has an async twin. Planned.

Parameters

LiveTennisMatchFetcher(api_key, status="live", tour=None, limit=10, base_url=None, timeout=30.0) — status/tour/limit can be overridden per run(), and run(match_id=...) fetches a single match. LiveTennisPlayerSearch(api_key, limit=10, base_url=None, timeout=30.0) — run(query, limit=None).

Development

pip install -e . pytest ruff
pytest                    # unit tests, fully mocked, no network
ruff check src tests examples
LIVETENNISAPI_KEY=... pytest -m integration   # live tests, needs a key

License

livetennisapi-haystack is distributed under the terms of the MIT license.