bolder-ai

The Python SDK for BEval Studio — log every LLM, VLM, and agent call in your stack to your BEval dashboard.


pip install bolder-ai


import beval
 
beval.init()
 
beval.log(
    kind="llm",
    model_id="gpt-4o-mini",
    input="What is the capital of France?",
    output="Paris.",
    latency_ms=312,
    tokens_in=7,
    tokens_out=2,
)

Distribution name: bolder-ai. Import name: beval.

What it does

Captures every call your app makes to an LLM or VLM, and every agent turn
Ships to BEval — input, output, model, latency, tokens, cost, errors, images
Fire-and-forget — logging never blocks your code, never raises on network errors
Zero lock-in — auto-wrappers for OpenAI and Anthropic, or log manually

Three ways to use it

	How	When to pick
`beval.log(...)`	Call directly from your code	Maximum control. Custom pipelines. Non-standard providers.
`beval.wrap(client)`	Monkey-patch an OpenAI or Anthropic client	Fastest integration. Covers 90% of apps.
`@beval.trace`	Decorator on any function	Agent loops, tool calls, custom orchestration.

You can mix all three in the same app.

Next steps

Quickstart — ship in five minutes
Installation — pip install options and optional extras
Concepts — what a log is, log kinds, projects and tenants
API Reference — every public function and field