Verified evidence, three method calls.
The Skippy Python SDK is in private preview. The verification and audit tooling that powers Skippy internally is available now as open-source packages.
Verification tooling, available now
The packages below are the verification, audit, and key infrastructure used inside Skippy. All are Apache 2.0, installable via pip, and require no Skippy account or API key.
skippy-verifyAvailableOffline cryptographic verifier for signed Skippy responses. Checks Ed25519 signatures, Merkle root integrity, key revocation, and optional Rekor transparency-log inclusion. No Skippy infrastructure required after install.
skippy-verifierAvailableClaim-vs-source entailment audit harness. Extracts claims from output, validates each against cited sources using pluggable NLI backends, and returns structured violation reports with JSON Schema-validated audit logs.
skippy-keysAvailablePublic key registry for cryptographic verification of signed Skippy responses. Provides key lookup, TTL-cached resolution, and key revocation status. Required by skippy-verify to resolve the signing key embedded in a signed response.
Bring your own entailment model. Any class implementing the NLIProvider Protocol works as a drop-in backend — no inheritance required.
skippy-python — verified evidence, three method calls
- verify() — claim verification with calibrated confidence
- retrieve() — evidence retrieval by entity
- score_output() — per-sentence grounding scores
- Pydantic models for all responses
- Typed error classes
- LangChain integration
- LlamaIndex integration
- OpenAI function-calling adapter
- Async support
- Batch verification
- Streaming responses
- Local cache
- Offline mode
Request SDK access
Tell us about your use case and we'll provision early access keys plus integration support.