Keep agent retrieval current as documents change with Serviette
Configure a watched document corpus, serve source-backed retrieval to agents, and verify that additions, edits, and deletions reach the index without manual re-ingestion.
Prerequisites
Python 3.10+; pip 23.1+; Serviette CLI; required Pathway license key; selected embedding backend and vector store; HTTP-capable agent
Installation
Install or set up from the source-backed instructions:
Create an isolated Python environment, use pip 23.1+, and run python -m pip install 'serviette[local]'. Obtain a free Pathway license key through the upstream license page and set PATHWAY_LICENSE_KEY. Run serviette wizard, choose the document folder, local embeddings and DuckDB, and then serviette up --config config.yaml. Use the upstream CPU PyTorch wheel option if appropriate for the host. Configure optional model credentials only if generated answers or remote embeddings are needed. Query POST /api/v1/retrieve with query and k; test a controlled file edit and deletion before relying on freshness.