A local, Postgres-backed long-term memory substrate for AI assistants. It ingests the conversational sediment of a working life and exposes it back as searchable context, entirely on your own hardware.
Assistants forget. Every conversation starts from nothing, and the useful context lives scattered across meeting transcripts, old chat exports, documents, and recordings that nobody is ever going to re-read.
Reliquary is the layer underneath that. Anything dropped into its inbox gets picked up, parsed, indexed, and made queryable: meeting transcripts, Claude and ChatGPT conversation exports, source code, configs, Markdown, HTML, Word documents, PDFs, email files, images through OCR, and audio through local GPU transcription.
Retrieval is hybrid rather than pure vector search: semantic nearest-neighbour and full-text search run side by side and get fused by reciprocal rank, because the thing you half-remember is sometimes a concept and sometimes an exact string. The whole surface is exposed to assistants over the Model Context Protocol.
Reliquary is an internal Locke Werks product, built single-user and local-first. The source is private and there is no public download. If you were hoping for a dashboard, this is the wrong product.
For access, licensing, or a conversation about running something like this in your own environment, write to archon@lockewerks.com.