feat-sdk 0.3.1

feat feature-flag SDK for Python (server-side, local evaluation)

---

feat Python SDK

Server-side Python SDK for feat feature flags. Local flag evaluation against a polled datafile. Zero runtime dependencies (stdlib only).

Install

pip install feat-sdk

Python 3.10+.

Usage

from feat import Client, ClientConfig, EvalContext

client = Client(ClientConfig(
    api_key="feat_sdk_...",
    url="https://data-01.feat.so",  # optional; this is the default
))
client.ready()

ctx = EvalContext(
    targeting_key="user-123",
    kinds={"user": {"plan": "pro", "email": "[e-mail]"}},
)

if client.get_boolean_value("checkout-v2", False, ctx):
    # ...
    pass

client.close()

Use a server API key (feat_sdk_...).

How it works

• Fetches a per-environment datafile and keeps it in memory.
• Streams live updates by default over Server-Sent Events: a background thread holds the connection and applies each pushed datafile the moment it changes.
• A background poll runs alongside the stream as a safety net. Its cadence follows stream health: while the stream is connected the poll runs at a slow 10-minute interval (the stream is the live path, so the poll only guards against a silently wedged connection); when streaming is off or the stream drops, the poll reverts to the normal interval (poll_interval_seconds, 30s by default and configurable) and becomes the primary refresh path. A dropped stream switches back to the fast interval immediately. Polls are ETag-aware via If-None-Match, and the stream reconnects with exponential backoff.
• Updates are version-ordered: a datafile is adopted only when its version is strictly newer than the one in memory, so the stream and poll never clobber each other.
• Evaluation runs in-process: no per-flag network call.
• close() stops the stream and poll threads cleanly.

Disabling streaming

Set streaming=False to rely on polling alone:

client = Client(ClientConfig(api_key="feat_sdk_...", streaming=False))

License

MIT