Skip to content

Voidloom 1.0.0

Latest
Latest

Choose a tag to compare

View all tags
@atg221 atg221 released this 27 Oct 13:42
· 4 commits to main since this release
v1.0.0
11c96bf

Voidloom 1.0.0 — the first public release. A living memory tapestry woven for AI agents.

Voidloom gives an AI coding agent a memory that lasts between sessions. You (or the agent) store what was learned and recall it later by meaning. A session-start hook hands Claude Code or Codex its relevant memories before the first prompt. Everything runs on your own machine: one command-line tool, one hook, and one SQLite file that is the brain.

Highlights

  • Local-first. Storing and recalling never leave your machine. No account, no telemetry.
  • Recall by meaning. Vector and keyword search fused by reciprocal rank, with an optional reranker; a keyword-only engine that needs nothing but Python is always there.
  • Starts competent. A fresh brain is loaded with 82 memories that teach an agent how to use it.
  • Claude Code and Codex. A session-start hook for both, and an instruction block for tools without hooks.
  • Maintains itself, if you want it to. Optional scheduled jobs consolidate, link, resurface and review memories. They are off until you turn them on, and the model they think with (or none) is your choice.
  • Honest installer. It says which compute it chose and how proven that is, previews everything with --dry-run, lists every path it may touch, asks rather than guesses, and finishes with a self-check. uninstall.sh removes it again and keeps your memories unless you ask otherwise.

Get started

git clone https://github.com/void-wright/voidloom.git
cd voidloom
bash install.sh

Then follow docs/GETTING_STARTED.md. Proven on Linux (including WSL2) with or without an NVIDIA GPU, and on Apple Silicon Macs with the GPU or the CPU. Intel Macs and AMD or Intel GPUs are supported but not yet proven on real hardware (see the README).

Found while preparing this release

  • recall --top-k K now returns the K most relevant memories (the diversity step used to drop every second-ranked result).
  • A fresh clone installs with plain bash install.sh.
  • Machines whose Python cannot load sqlite_vec now get their vector tables.

The full list, and the known issues, are in CHANGELOG.md. Benchmark results on standard memory benchmarks are in preparation.

Licensed under the Apache License 2.0.

Assets 2
Loading