krepis (Greek κρηπίς — the foundation course a structure stands on) is a small, typed library of general-purpose Python primitives for building production data and LLM pipelines on AWS.
pip install krepis| Module | Purpose |
|---|---|
logging |
Structured logging + the flow-doctor secret-seeding chokepoint |
secrets |
SSM-backed secret resolution |
alerts |
Telegram / SNS alert routing with S3-marker dedup (incl. CLI) |
telegram |
Telegram transport |
email_sender |
SES / SMTP email transport |
http_retry |
Bounded-backoff HTTP retry with jitter |
locks |
S3 conditional-PUT writer locks |
cost |
LLM cost telemetry (token counts × price card → USD) |
model_metadata |
ModelMetadata value object (per-invocation model + token cost) |
anthropic_payload |
Anthropic request-payload construction + validation |
metrics |
Typed metric records |
trading_calendar |
NYSE trading-day calendar (pure stdlib) |
dates |
Calendar / trading-day dual-track date helpers |
ec2_spot |
EC2 spot-instance resilience helpers |
ssm_dispatcher |
SSM command dispatch |
ssm_log_capture |
SSM run-command log capture |
usage_pacing |
Linear-pace short-circuit against a rolling reset-window quota |
MIT — see LICENSE and NOTICE. Contributions accepted
under the DCO (git commit -s).
Part of the Nous Ergon project.