Foresportia Python SDK

Documentation · API docs

Official Python SDK for the Foresportia API, a football analytics API that provides match data, model probabilities, predicted picks, confidence signals, and Core Analytics payloads for building or enriching football prediction models.

The SDK targets server-side Python (3.9+) with a synchronous, typed client.

Public plans share the same SDK surface:

• Developer — free: 1 competition, up to 7 days of verified history, bulk requests up to 5 match IDs, and a limited analytics projection.
• Starter — paid: 20 competitions, up to 90 days of verified history, bulk requests up to 100 match IDs, and more complete analytics.
• Legacy beta: temporary compatibility for existing keys.

Installation

pip install foresportia

Optional extras:

pip install "foresportia[ml]"    # numpy + scikit-learn for the ML example
pip install -e ".[dev]"          # local development from a clone

Authentication and key safety

Authentication uses the X-API-Key header only. The SDK never puts the key in a URL, and the key never appears in repr(), logs, or exception messages.

Store the key in an environment variable rather than in source code:

export FORES_API_KEY="fs_developer_your_key_here"

$env:FORES_API_KEY = "fs_developer_your_key_here"

from foresportia import ForesportiaClient

client = ForesportiaClient.from_env()          # reads FORES_API_KEY
# or explicitly:
client = ForesportiaClient(api_key="...", timeout=10.0)

The base URL must be HTTPS. Plain HTTP is only accepted for localhost or with allow_insecure_base_url=True (development/testing only).

Do not commit API keys to source control, notebooks, screenshots, or support tickets.

Quickstart

from foresportia import ForesportiaClient

with ForesportiaClient.from_env() as client:
    leagues = client.list_leagues()
    for league in leagues.data:
        print(league.code, league.name, league.matches_available)

    matches = client.list_league_matches(
        "PREMIER_LEAGUE",
        include="upcoming",
        start="2026-07-15",
        days=7,
    )
    for match in matches.data:
        print(match.kickoff, match.home_team, "vs", match.away_team, match.pick)

The SDK also accepts Starter and legacy beta keys without inferring a plan from their prefix. The server remains authoritative for entitlements.

Every typed method returns an ApiResponse with:

response.data           # typed result (list[League], list[MatchSummary], MatchDetail, BulkResult)
response.payload        # full JSON payload as a dict
response.etag           # ETag header for conditional requests
response.quota          # parsed rate-limit / quota headers
response.status_code    # HTTP status
response.not_modified   # True for 304 responses

Available competitions

leagues = client.list_leagues()
for league in leagues.data:
    print(league.code, league.name, league.country, league.activity_status)

Your API key only sees the competitions enabled for it. Use the codes exactly as returned (for example in list_league_matches).

Match detail

Match IDs from list endpoints look like fsm:v1:<64 hex characters>. Pass them exactly as returned:

match = client.get_match("fsm:v1:0123...abcdef")
detail = match.data
print(detail.home_team, "vs", detail.away_team)
print(detail.probabilities)   # 1X2, over/under, BTTS, draw no bet, ...
print(detail.ratings)         # ELO-style ratings
print(detail.raw)             # complete Core Analytics payload

Bulk limits by plan

Developer accepts up to 5 match IDs per request; Starter accepts up to 100. The SDK validates only the technical ceiling of 100. A Developer request with more than 5 IDs is sent and the server returns bulk_limit_exceeded. Order is preserved and per-ID failures do not hide successful results:

bulk = client.get_matches_bulk([id_1, id_2, id_3])
for detail in bulk.data.results:
    print(detail.id, detail.home_team, "vs", detail.away_team)
for error in bulk.data.errors:
    print("failed:", error.match_id, error.code)   # e.g. match_not_found

History and pagination

Developer provides up to 7 days of verified history and Starter up to 90 days. Each request has a maximum window of 31 days. The effective archive may start later at response.history_available_from while it fills progressively.

page = client.list_league_matches("CHN", include="past", days=7, limit=50)
while True:
    for match in page.data:
        print(match.kickoff, match.home_team, match.away_team)
    if not page.next_cursor:
        break
    page = client.list_league_matches(
        "CHN", include="past", days=7, limit=50, cursor=page.next_cursor
    )

# Or stream rows without loading every page:
for match in client.iter_league_matches("CHN", include="past", days=7, limit=50):
    print(match.id)

The server may return history_window_exceeded when the requested dates fall outside the key's entitlement.

History helpers

list_league_history() and iter_league_history() are thin wrappers that always request finished matches (include="past"); they take the same parameters, return the same types, and expose the same metadata and errors as the methods above. match.is_final mirrors the server status == "final" and match.predicted_outcome returns the published pre-match pick ("home", "draw", "away", or None); result_score is the final score from the API. No live data is added.

By default days is omitted, so the server picks the window from the key's real history entitlement, capped at 31 days per request. Pass days explicitly to choose a shorter or specific window:

with ForesportiaClient.from_env() as client:
    # Automatic window: the available history, capped at 31 days per request.
    page = client.list_league_history("SUE")

    # Or choose an explicit window.
    page = client.list_league_history("SUE", days=3)

    for match in client.iter_league_history("SUE"):
        if not match.is_final:
            continue
        print(
            match.home_team,
            match.away_team,
            match.result_score,
            match.predicted_outcome,
        )

Developer keys currently expose 7 days of history; Starter keys have a 90-day rolling entitlement, but an automatic request stays capped at 31 days. iter_league_history() paginates the requested window — it does not automatically split a 90-day entitlement into several windows. The SDK never resolves the entitlement locally and makes no extra plan-discovery call; the server stays authoritative and a valid window may legitimately contain no matches.

Today endpoints

today = client.list_today_matches()
picks = client.list_today_picks(limit=20)

Public health

client.health() returns the tolerant dictionary from the stable public GET /v1/health endpoint. It reports service/data health and is not an entitlement or account-status check. Unknown future fields are preserved.

Quotas and rate limits

The API reports quota state in response headers; the SDK parses them into response.quota (fields are None when the header is absent):

response = client.list_leagues()
print(response.quota.remaining)                  # X-RateLimit-Remaining
print(response.quota.units_remaining_hour)       # X-Quota-Units-Remaining-Hour
print(response.quota.units_remaining_day)        # X-Quota-Units-Remaining-Day
print(response.quota.match_rows_remaining_day)   # X-Match-Rows-Remaining-Day

Quota enforcement stays server-side. On HTTP 429 the SDK raises ForesportiaRateLimitError (or ForesportiaConcurrencyLimitError) carrying retry_after and the quota snapshot. Retrying after 429 is opt-in:

client = ForesportiaClient.from_env(retry_on_rate_limit=True, max_retries=2)

ETags and conditional requests

Pass a previous ETag to skip unchanged payloads. A 304 Not Modified is a normal response, not an error:

first = client.get_match(match_id)
second = client.get_match(match_id, etag=first.etag)
if second.not_modified:
    detail = first.data      # reuse the cached payload
else:
    detail = second.data

Error handling

from foresportia import (
    ForesportiaAuthenticationError,   # 401
    ForesportiaAuthorizationError,    # 403
    ForesportiaNotFoundError,         # 404
    ForesportiaValidationError,       # 400 / 422 / client-side validation
    ForesportiaRateLimitError,        # 429
    ForesportiaConcurrencyLimitError, # 429 concurrency_limit_exceeded
    ForesportiaServerError,           # 5xx
    ForesportiaTransportError,        # network / timeout
    ForesportiaAPIError,              # base class for all of the above
)

try:
    matches = client.list_today_matches()
except ForesportiaRateLimitError as exc:
    print("rate limited, retry after", exc.retry_after, "seconds")
except ForesportiaAPIError as exc:
    print(exc.status_code, exc.error_code, exc.endpoint)

Exceptions carry status_code, error_code (the API business code such as match_not_found or bulk_limit_exceeded), endpoint, retry_after, and a quota snapshot when available. The API key is never included.

Common business codes include competition_not_selected, history_window_exceeded, and bulk_limit_exceeded. starter_required is normally an availability value for a masked Developer field; if a future endpoint returns it as an error code, it remains available through exception.error_code. Unknown future codes use the same safe fallback.

Retries and timeouts

• Explicit timeout per client (timeout=10.0, default 20 s).

• Retries are disabled by default (max_retries=0). A request that times out client-side may still have been processed and counted against your quota server-side, so each retry can consume an extra quota unit. Opt in explicitly when that trade-off is acceptable:

  client = ForesportiaClient.from_env(max_retries=2)
  

  When enabled, only GET requests are retried, on network errors and 502/503/504, with bounded exponential backoff.

• No automatic retry on 400/401/403/404, nor on 429 unless retry_on_rate_limit=True.

• The bulk POST is never retried on 5xx.

• Use the client as a context manager (with ... as client:) or call client.close().

Machine learning example

A runnable script is provided in examples/ml_starter_features.py: it fetches upcoming matches, extracts probabilities, markets, and confidence into a feature matrix, and optionally fits a small scikit-learn model (pip install "foresportia[ml]"). Foresportia outputs are model probabilities, not guaranteed predictions; you need your own historical labels to train anything meaningful.

rows = []
for match in client.list_league_matches("PREMIER_LEAGUE", days=14).data:
    p = match.probabilities
    rows.append([p.get("home"), p.get("draw"), p.get("away"),
                 match.markets.get("btts"), match.confidence.get("score")])

Legacy methods

The v0.1 dict-based methods keep working unchanged:

client.me()
client.usage()
client.picks_today()
client.matches_today()
client.leagues()
client.league_matches("FIN", include="all", days=14, limit=10)
client.world_cup_2026_matches(limit=10)

Prefer the typed list_* / get_* methods for new code.

Links

• API documentation: https://www.foresportia.com/api/docs/
• LLM-friendly API reference: https://www.foresportia.com/api/docs/llms-full.txt
• Homepage: https://www.foresportia.com
• Developer docs (EN): https://www.foresportia.com/en/developers.html
• API dashboard (EN): https://www.foresportia.com/en/api-dashboard.html
• Repository: https://github.com/QBarbedienne/foresportia-python
• Issues: https://github.com/QBarbedienne/foresportia-python/issues

Disclaimer

Foresportia provides football probabilities, model outputs, and analytics data. It does not provide betting advice, financial advice, bookmaker odds, guaranteed outcomes, or instructions to place wagers. Any use of Foresportia data is the responsibility of the user and should comply with applicable laws, regulations, and platform policies.

Starter prediction reliability (since 0.3.2)

Use SDK 0.3.2 or later and a Starter key with an API deployment supporting include_unreliable. Upgrade with python -m pip install --upgrade foresportia.

Typed match methods accept include_unreliable=True to explicitly request complete, verified calculations rejected for identified sporting quality reasons. By default the API still returns reliable predictions only. This does not extend the match horizon.

response = client.list_league_matches("LIGUE_1", include_unreliable=True)
for match in response.data or []:
    print(match.id, match.reliability, match.reliability_context)

Reason codes: elo_unreliable, shadow_insufficient_data, sparse_goal_context, elo_missing, goal_raw_data_unreliable, uefa_qualification_mute. These flags are not a probability of correctness. Invalid calculations and unknown rejection reasons remain excluded; the original probabilities are unchanged. No translated message is returned. Keep the option unchanged when paginating.

License

MIT. See LICENSE.