What does provenance-aware priority handling do?

It prevents untrusted RAG, tool, and assistant content from gaining protected priority solely through imperative or safety-critical wording.

Does LLMSlim prevent prompt injection?

No. It mitigates compression-induced instruction elevation and applications still need defense in depth.

The default extractive strategy runs locally. Rewrite and hybrid strategies use a caller-supplied provider.

API Reference 4 min readUpdated August 2026

Use explicit provenance for retrieval and conversational context.

compress_documents() defaults to ContextRole.RAG. This is an intentional v0.3.1 behavior change.

rag.py

from llmslim import compress_documents
results = compress_documents(documents, query=user_query, target_ratio=0.4)

compress_chat_messages() propagates system, developer, user, assistant, and tool roles. Unknown roles map to GENERAL.

chat.py

from llmslim import compress_chat_messages
compressed = compress_chat_messages(messages, target_ratio=0.5)

Frequently asked questions

Are system messages compressed by default?

No. The default chat pipeline leaves system messages unchanged.