Loop Engine — Perplexity Sonar adapter and guard evidence APIs
Ships Perplexity Sonar as an LLM adapter with citations and bounded retries, plus core LLMAdapter I/O contracts and deep guardEvidence() for redacting sensitive evidence in audit outputs.
What's new
Perplexity Sonar LLM adapter
@loop-engine/adapter-perplexity implements LLMAdapter with Sonar citations, bounded retries, and guardEvidence integration for audit-safe outputs.
Core adapter contract and evidence redaction
@loop-engine/core adds LLMAdapter, AdapterInput, AdapterOutput, and deep guardEvidence() so adapters can attach structured evidence without leaking sensitive data in logs or audit trails.
Documentation
docs/integrations-perplexity.md explains Sonar vs Perplexity Computer skills and when to use this adapter.
Why it matters
Security and compliance teams need LLM steps to be inspectable: citations for grounding, and predictable redaction when evidence must not leave the audit trail.
Shipping Sonar behind the same LLMAdapter and guardEvidence path as other providers means you can add research-style models without forking your governance model.
This targets teams automating operational loops where "the model said so" is not an acceptable control. You need evidence objects and policy-friendly logs.