Python

bryel ships your model calls, tool calls, tokens, and cost to bryel as OpenInference traces. One call configures OpenTelemetry; then any OpenInference instrumentor flows through — bryel adds no mapping, it’s just the exporter.

Install

uv add bryel

pip install bryel

Initialize

import bryel
bryel.init(api_key="bk_…", service_name="my-app")

Or read the key from the environment (BRYEL_KEY) and omit api_key.

Instrument your stack

Add the OpenInference instrumentor for your library:

from openinference.instrumentation.openai import OpenAIInstrumentor
OpenAIInstrumentor().instrument()

from openinference.instrumentation.anthropic import AnthropicInstrumentor
AnthropicInstrumentor().instrument()

from openinference.instrumentation.langchain import LangChainInstrumentor
LangChainInstrumentor().instrument()

from openinference.instrumentation.llama_index import LlamaIndexInstrumentor
LlamaIndexInstrumentor().instrument()

…or let bryel wire up every instrumentor you have installed:

bryel.init(api_key="bk_…", instrument=True)

Verify

Run a model call, then open the trace explorer — model, tokens, cost, and tool calls appear.

Full example

import bryel
bryel.init(service_name="support-agent", instrument=True)  # reads BRYEL_KEY

from openai import OpenAI
OpenAI().chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "hello"}],
)

bryel.shutdown()  # flush pending spans before exit

Grouping and feedback

Stamp session.id, user.id, and a bryel.interaction.id you mint onto your spans to group turns into a conversation and join feedback.

bryel sends OTLP/HTTP protobuf and needs no client-side mapping — the ingest normalizes both the OpenInference (llm.*) and OpenTelemetry GenAI (gen_ai.*) conventions.

Evals

Ship an eval run’s artifacts (screenshot and/or text) for judging — one eval task = one session_id. See the evals guide for the full flow.

import bryel

session = bryel.start_eval_session(session_id, api_key)
# …run your agent on the case prompt, capture a screenshot…
session.export(images=[screenshot_base64], output_texts=[final_answer])

API

`bryel.init(api_key=None, *, service_name=None, endpoint=None, headers=None, set_global=True, instrument=False)`

Configures the exporter and returns the OpenTelemetry TracerProvider.

str

Project API key (bk_…). Falls back to $BRYEL_KEY.

str

OpenTelemetry service.name.

str

default:"https://ingest.eu.bryel.ai/v1/traces"

OTLP/HTTP URL. Override via $BRYEL_INGEST_URL for self-hosting.

dict

Extra headers, merged after auth.

bool

default:"True"

bool

default:"False"

Also auto-instrument installed OpenInference libraries.

`bryel.auto_instrument(tracer_provider=None) -> list[str]`

Best-effort instrument every installed OpenInference library — provider SDKs, agent SDKs, and frameworks (see Integrations for the full list). Returns the ones it wired up; safe to call with none installed.

`bryel.shutdown()`

Flush pending spans and shut down. Call on process exit.

`bryel.start_eval_session(session_id, api_key, *, endpoint=None) -> EvalSession`

Bind (api_key, session_id) and return a handle whose export(images=…, output_texts=…) ships the run’s artifacts. See evals.

`bryel.export_eval_session(api_key, session_id, *, images=None, output_texts=None, endpoint=None) -> dict`

Ship an eval run’s artifacts directly. images are base64 strings; provide images and/or output_texts.

Getting started

SDK

Concepts

Guides

Full example

Grouping and feedback

Evals

API

`bryel.init(api_key=None, *, service_name=None, endpoint=None, headers=None, set_global=True, instrument=False)`

`bryel.auto_instrument(tracer_provider=None) -> list[str]`

`bryel.shutdown()`

`bryel.start_eval_session(session_id, api_key, *, endpoint=None) -> EvalSession`

`bryel.export_eval_session(api_key, session_id, *, images=None, output_texts=None, endpoint=None) -> dict`

​Full example

​Grouping and feedback

​Evals

​API

​bryel.init(api_key=None, *, service_name=None, endpoint=None, headers=None, set_global=True, instrument=False)

​bryel.auto_instrument(tracer_provider=None) -> list[str]

​bryel.shutdown()

​bryel.start_eval_session(session_id, api_key, *, endpoint=None) -> EvalSession

​bryel.export_eval_session(api_key, session_id, *, images=None, output_texts=None, endpoint=None) -> dict

Full example

Grouping and feedback

Evals

API

`bryel.init(api_key=None, *, service_name=None, endpoint=None, headers=None, set_global=True, instrument=False)`

`bryel.auto_instrument(tracer_provider=None) -> list[str]`

`bryel.shutdown()`

`bryel.start_eval_session(session_id, api_key, *, endpoint=None) -> EvalSession`

`bryel.export_eval_session(api_key, session_id, *, images=None, output_texts=None, endpoint=None) -> dict`