Does LLMSlim process multimodal inputs?

No. The released engine processes text.

LLM ProviderPython integration

LLMSlim + Google Gemini

Compress retrieved Python context before a Gemini SDK request.

Use compress_documents() for retrieved text and preserve accurate provenance.

Install LLMSlim and the official Google Gemini SDK using your package manager.

terminalbash

pip install llmslim

1. Retrieve or construct application context.

2. Label provenance accurately and compress with LLMSlim.

3. Send the resulting text to the provider SDK.

4. Apply application-level policy and output validation.

gemini_app.pypython

from llmslim import compress_documents

compressed = compress_documents(documents, query=query, target_ratio=0.4)

Use LLMSlim in your Python application before constructing the provider request. Measure target workloads before choosing a compression ratio.

Use compress_documents() for retrieved documents; it defaults to RAG provenance.
Leave system messages unchanged by default in compress_chat_messages().
Do not treat compression as a complete prompt-injection defense.

Metric	Uncompressed	LLMSlim	Impact
v0.4.0 release gate	N/A	489 tests passed	90.93% coverage; 375 schemas across 18 catalogs

Unexpected compression result

Check target_ratio, role labels, and whether the input is long enough to benefit from compression.