Braintrust alternative.
Evals + replay, with a real gateway underneath.

from braintrust import init_logger
from openai import OpenAI

init_logger(project="my-proj")

client = OpenAI(
    base_url="https://api.braintrust.dev/proxy/v1",
    api_key="bt-...",
)

r = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{"role": "user", "content": "Hello"}],
)

from openai import OpenAI

client = OpenAI(
    base_url="https://api.aigateway.sh/v1",
    api_key="sk-aig-...",
)

r = client.chat.completions.create(
    model="openai/gpt-5.4",
    messages=[{"role": "user", "content": "Hello"}],
)

• Evals are a gateway primitive — eval results bind to a routing alias directly.
• Every modality, not just text LLM evals.
• Replay + shadow A/B on any past request, zero extra wiring.
• Sub-account API, cost tags, caching — an actual production gateway, not just eval infra.
• Pay 5% at top-up; no seat tax.

• Braintrust has a richer eval-authoring UI for complex rubrics; we match on programmatic + CLI flows.