replicase 0.1.0

Trust infrastructure for scientific AI agents — trace, cite, score.

replicase

Trust infrastructure for scientific AI agents.

import replicase

replicase.init(api_key="rpl_test_...")

@replicase.trace(name="my_research_agent")
async def search(query: str) -> str:
    # ... your agent code ...
    replicase.cite(doi="10.1038/...", title="...", claim="X causes Y")
    return answer

Every traced call is captured, citations are NLI-scored against their source, and results show up in the Replicase dashboard. See the parent repo for full docs.

Install

pip install replicase

Configuration

Env var	Default	Required	Notes
REPLICASE_API_KEY	None	Yes	Your project key
REPLICASE_ENDPOINT	https://api.replicase.bio	No	Override for local dev

Privacy

@replicase.trace captures function inputs and outputs and ships them to the API. Built-in regex redaction (default-on) scrubs common API-key prefixes, JWTs, bearer tokens, emails, US-SSN-shaped, and 13–19 digit card-shaped substrings. This is best-effort, not a security boundary — base64-encoded secrets, custom-format tokens, and internal hostnames will not be caught. Scrub at the source for regulated data.

Disable redaction with replicase.init(redact=False) (not recommended in prod) or supply a callable: replicase.init(redact=my_redactor).