MemPalace Extended
Repository name: mempalace-extended
MemPalace Extended is a GitHub-ready extension of the original
MemPalace project.
It keeps the original local-first memory workflow and adds:
- read-only ingestion for
PDF, DOCX, and XLSX
- image asset storage for
PNG, JPG, JPEG, GIF, WEBP, and BMP
- mixed
text + image MCP responses
- folder watch with
start / stop / status
- stable embedding cache configuration
The Python package and CLI command remain mempalace for compatibility.
Attribution
This repository is based on the original MemPalace project by
milla-jovovich and contributors.
- Original repository: https://github.com/milla-jovovich/mempalace
- Original license: MIT
- Upstream source baseline used for this extension: original
main branch
source version 3.0.14
- Original repository's latest visible GitHub Release:
v3.0.0
- Extension status: this repository adds document ingestion, image asset
handling, MCP multimodal responses, and folder watching on top of the
original project structure
See NOTICE.md and LICENSE.
Upstream Baseline Note
This extension was prepared from the upstream source state where
mempalace/version.py reported 3.0.14.
That is different from the original repository's latest visible GitHub Release
tag, which is currently v3.0.0.
For provenance, the most accurate short description is:
based on original MemPalace main-branch source version 3.0.14
What This Version Focuses On
This extension is intended for practical local knowledge management:
- mine source code, notes, reports, papers, and spreadsheets into one palace
- keep source directories untouched during indexing
- let MCP-compatible AI clients search the palace and retrieve related images
- monitor a folder continuously and ingest new knowledge automatically
This repository does not try to preserve the original README's benchmark and
marketing framing. It documents the extension as a usable engineering fork.
Feature Summary
Core memory workflow
mempalace init <dir>mempalace mine <dir>mempalace search "query"mempalace statusmempalace wake-up
Extended project ingest
- plain text, Markdown, code, JSON, YAML, CSV, SQL, HTML, CSS, TOML
PDF text extraction via pypdfDOCX text and table extraction via python-docxXLSX sheet and cell extraction via openpyxl- image asset indexing with sidecar metadata and palace-managed copies
Extended MCP behavior
- semantic text search
- image asset retrieval through
mempalace_get_asset
- search responses can include related images alongside text hits
- works with MCP clients that support
stdio servers
Automatic folder watch
mempalace watch start <dir>mempalace watch statusmempalace watch stop <dir-or-id>- supports project folders and conversation export folders
- optional pruning of drawers for deleted files
Supported Inputs
Projects mode
Supported directly:
.txt, .md.py, .js, .ts, .jsx, .tsx.json, .yaml, .yml, .toml.html, .css.java, .go, .rs, .rb, .sh, .sql, .csv.pdf, .docx, .xlsx.jpg, .jpeg, .png, .gif, .webp, .bmp
Conversations mode
Supported directly:
Built-in normalization handles:
- Claude Code JSONL
- OpenAI Codex CLI JSONL
- Claude exports
- ChatGPT conversations JSON
- Slack exports
Runtime Environment
Compatibility at a glance
| Target | Python | Status | Notes |
|---|
| Windows local use | 3.11 | recommended | best default for conda or uv |
| Windows local use | 3.10 or 3.12 | supported with caution | often workable, but not the primary local baseline |
| Windows local use | 3.13 or 3.14 | not recommended | may require native builds in the chromadb dependency chain |
| Linux or macOS | 3.9, 3.11, 3.13 | CI-validated | current GitHub Actions matrix runs on Ubuntu |
Recommended baseline
- operating system: Windows 10 or Windows 11
- shell: PowerShell or Windows Terminal
- Python:
3.11
- environment manager:
conda or uv
- storage: local SSD with enough space for the palace, image assets, and embedding cache
- MCP client: a client that supports
stdio MCP servers and tool calling
Supported Python range
