boei 0.1.0

Boei runtime SDK and OpenTelemetry instrumentation for AI systems

Boei Runtime SDK

The Boei Python SDK captures runtime metadata for AI systems and sends it to Boei's runtime ingest API. Install it:

pip install boei

Then import and initialize:

import boei

boei.init()

Activation

Use a managed project key from Boei:

export BOEI_API_KEY="boei_pk_..."

The SDK exchanges the project key for a runtime token at /v1/auth/token and sends telemetry to /v1/runtime/ingest or /v1/traces. There is no Boei CLI in this story; activation is SDK install plus project key configuration.

Privacy Default

Boei does not read prompt or response content by default.

export BOEI_CAPTURE_CONTENT=false

That is also the default when the variable is unset. The legacy TRACELOOP_TRACE_CONTENT variable does not turn Boei content capture on.

If a tenant deliberately opts in:

export BOEI_CAPTURE_CONTENT=true

the SDK still runs redaction for common PII and secret patterns before export. The ingest API also strips prompt/message attributes unless the runtime token carries capture_content=true.

What Runtime Proves

Runtime evidence can corroborate:

• call volume
• token counts
• provider
• model choice
• SDK-estimated runtime cost under a project key

Runtime evidence cannot prove:

• provider bill-of-record spend
• system ownership
• a confirmed AI system
• control effectiveness
• prompt or response content when content capture is off

Runtime rollups become source_evidence_atoms as pending key observations. They stay corroborator evidence until later scanner binding and reviewer confirmation work.