memquilt 0.0.0

Federated memory layer for AI tools. Stitches AI memories across tools instead of replacing them. Pre-alpha placeholder — active development.

MemQuilt (Python)

The federated memory layer for AI. Stitches, doesn't replace.

MemQuilt stitches AI tools' memories into one federated layer. Each tool keeps its own store — MemQuilt adds cross-tool search, knowledge graph, and dreaming on top. Open source. Local-first.

Website: https://memquilt.com Repository: https://github.com/memquilt/memquilt

---

Status: Pre-alpha placeholder

This package (version 0.0.0) is a namespace placeholder. The functional implementation is actively under development. The first working release (0.1.0) is expected within the next 3–6 months.

If you install this today, you will get:

import memquilt

print(memquilt.hello())
# MemQuilt — stitches AI memories across tools. Pre-alpha placeholder. See https://memquilt.com

That's all for now. Watch the repository for updates.

---

What MemQuilt will be

MemQuilt is a dual-layer memory system for AI tools:

1. Backend layer — gives "dumb" tools (Codex, Cursor, Gemini CLI, bare shells) a real memory backend with Weibull decay, three-tier promotion, five-state LLM dedup, and threat scanning.
2. Federation layer — lightweight pointer index over "smart" tools that already have their own memory (OpenClaw, Hermes Agent, MemPalace, Claude Code). MemQuilt does not migrate their data; it only indexes and connects.
3. Global brain layer — unified cross-source scoring, knowledge graph (entities + temporal triples), and three-phase Dreaming (light / deep / REM) that reads both layers and writes consolidated insights back into the Backend layer only.

Design principle: don't replace source memory systems — stitch them into something greater. Like patches in a quilt, each memory source keeps its color, its cadence, its lineage. The golden seams between them are where cross-tool intelligence emerges.

---

License

MIT — see LICENSE.