Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3,128 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FathomDB

FathomDB is a local-first retrieval and graph-oriented data system for application and agent workloads. It embeds SQLite (FTS5 + sqlite-vec) and ships one engine behind three SDKs — Python, TypeScript and Rust — plus an operator CLI.

  • Hybrid retrieval: a vector branch and an FTS5 branch fused by Reciprocal Rank Fusion, with an optional CPU cross-encoder rerank and an optional graph-BFS third arm.
  • Governed, allowlisted command surface with a single typed error taxonomy shared 1:1 across the bindings.
  • Bitemporal-ish record model: transaction-time supersession, a world-time validity window on nodes, and edge t_valid / t_invalid.
  • Lifecycle and erasure verbs in the SDK — transition, purge, erase_source — so a consumer with no CLI on PATH can still discharge a deletion obligation.
  • Optional in-process default embedder (bge-small-en-v1.5, pure Rust).

Status: 0.8.20, pre-1.0 beta. The surface may change between micro releases. 0.8.20 has not been published to crates.io / PyPI / npm; publishing is a separate, gated step. Licensed MIT (see LICENSE).

Public documentation: docs/ (built with mkdocs build --strict). Changes since 0.8.9: CHANGELOG.md.

Repository layout:

  • docs/ contains public MkDocs source and client-facing technical positions.
  • dev/ contains internal engineering material: requirements, architecture, ADRs, subsystem design, interface contracts, and planning notes.
  • src/ contains implementation roots and unit-test-adjacent code.
  • test/ contains cross-language, smoke, fixture, and performance assets that are not package-local unit tests.

Implementation roots:

  • Rust workspace members live under src/rust/crates/
  • Python package root lives under src/python/
  • TypeScript package root lives under src/ts/

Start here:

  • Agent instructions: AGENTS.md — the canonical operating manual for AI coding agents; read it first if you are one
  • Public docs: docs/index.md
  • Internal docs index: dev/README.md — and from there the release schedule-of-record and the live board under dev/plans/
  • Workspace checks: scripts/check.sh

Common commands:

cargo check --workspace
pip install -e src/python/
cd src/ts && npm install
mkdocs build --strict

About

Local datastore memory for AI agents — graph, vector, and full-text search.

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages