ranke-db

README

RankeDB

Everything is Knowledge — Knowledge is Everything.

A provenance database for LLM context management.

RankeDB inverts the usual relationship between knowledge graphs and provenance: the provenance DAG is the system, and the semantic knowledge graph is a view projected from it. Nothing is ever deleted. Multiple levels of detail are preserved in parallel — from the raw source artifact up to the semantic triplet — networked by provenance, and the strategy of retrieval and reasoning is left to the consumer.

RankeDB is a data structure and a set of invariants, not a deployment topology.

Architecture

Three storage levels behind a single API:

• L0 — Raw Sources. Immutable, content-addressable object store. The archive.
• L1 — Provenance DAG. Append-only graph of Thoughts, each derived from Records or other Thoughts by a worker. Node authority: full content, full-text search, vector embeddings, provenance edges.
• L2 — Semantic Index. Materialized view of L1 optimized for associative retrieval. Edge authority: semantic relations, cross-domain links, temporal validity. Every node and edge traces back to L1.

Papers

Drafts in papers/:

1. 01-rankedb — the database architecture, design philosophy, comparative positioning
2. 02-rankedb-workers — building a semantic graph from raw data
3. 03-rankedb-retrieval — chat, memory agents, the coordination problem
4. 04-rankedb-coordination — Stacker: attention allocation in multi-agent chat

License

MIT (code) · CC-BY-4.0 (papers)