enlight-chakra-observability 0.1.0

One-line LLM observability for Python — a thin wrapper over OpenLLMetry that exports OTLP traces to Enlight Chakra.

enlight-chakra-observability

One-line LLM observability for Python. A thin, vendor-neutral wrapper over OpenLLMetry (Traceloop, Apache-2.0) that exports OTLP traces to your Enlight Chakra deployment. The single lk_… ingest key (copied from the Enlight Chakra Instrument app page) encodes the OTLP endpoint and auth — no base64, no endpoint URLs, no "which key goes where".

Under the hood it's OpenLLMetry → plain OTLP. You are never locked in.

Naming: the distribution is enlight-chakra-observability and the import name is enlight_chakra_observability. (The earlier enlight-sdk / import enlight_chakra collided with an unrelated PyPI package, Holoviz lumen. import enlight_chakra still works as a deprecated alias when this package is installed.)

Install

# From source (works today — not yet on public PyPI):
pip install "git+https://github.com/esds/lumen.git#subdirectory=sdk/python"
#   or, from a local checkout:   pip install ./sdk/python
#   or, from the built wheel:    pip install enlight_chakra_observability-0.1.0-py3-none-any.whl

# Once published to PyPI / your private index:
pip install enlight-chakra-observability

Use

import enlight_chakra_observability as lumen

enlight_chakra.init(api_key="lk_...", app_name="checkout-api")   # once, at startup
enlight_chakra.identify(user_id="u_42", session_id="s_7")        # optional, per request

# your normal OpenAI / Anthropic / LangChain / LlamaIndex code is now auto-traced

For short scripts/demos, flush each span immediately:

enlight_chakra.init(api_key="lk_...", disable_batch=True)

To make traces visible in a multi-tenant Enlight Chakra, stamp your tenant as a resource attribute (pass-through to Traceloop), or set OTEL_RESOURCE_ATTRIBUTES=tenant.id=...:

enlight_chakra.init(api_key="lk_...", resource_attributes={"tenant.id": "tnt_shiftx"})

What you get

Every request becomes a nested span tree in Enlight Chakra — retrieval, prompt assembly, the model call, tool calls, agent steps — grouped by user and session. See it under Span Explorer, RAG & Retrieval, Agents & Tools, and Sessions.

The lk_ key

lk_<base64url(endpoint|public_key|secret_key)>. It packs the OTLP endpoint and your project's Basic-auth credentials into one token, so init() needs nothing else. It contains a secret — treat it like any API key (env var / secrets manager; never commit).

Publishing

python -m build              # → dist/enlight_chakra_observability-0.1.0-*.whl + .tar.gz
twine upload dist/*          # public PyPI  (or: twine upload -r <private-index> dist/*)

License

Apache-2.0.