mempalace

MemPalace GHCP R Fork

Local-first GitHub Copilot memory for workstations where R is available and Python application installs are not. This fork replaces the Python package and ChromaDB storage layer with an R package, JSONL drawers, and Rscript-driven GHCP hooks.

MemPalace keeps the core promise: store user/project memory as verbatim local text and return the original words. Nothing is sent to an external service by default.

Requirements

• R 4.2 or newer. The target workstation for this fork is R 4.5.2 with Rtools 4.5.
• R package jsonlite is required.
• R packages digest, shiny, and openssl are optional. digest improves ID hashes, shiny runs the local viewer, and openssl is reserved for future encrypted-backup parity.

Install

From this repository:

Rscript -e "install.packages(c('jsonlite', 'digest'), repos = 'https://cloud.r-project.org')"
Rscript -e "install.packages('.', repos = NULL, type = 'source')"

Run the CLI through Rscript:

Rscript -e "mempalace::mempalace_cli()" --version
Rscript -e "mempalace::mempalace_cli()" mine . --wing mempalace_ghcp_r
Rscript -e "mempalace::mempalace_cli()" search "prior modeling decision"

GHCP Quickstart

Install the user-level Copilot hooks, skills, instructions, settings snippet, and R Shiny viewer:

powershell -ExecutionPolicy Bypass -File .\integrations\github-copilot\scripts\Install-MemPalaceGHCP.ps1 -RunDoctor -Force

Core GHCP commands:

Rscript -e "mempalace::mempalace_cli()" ghcp init --force
Rscript -e "mempalace::mempalace_cli()" ghcp compile-memory-pack
Rscript -e "mempalace::mempalace_cli()" ghcp search "prior modeling decision"
Rscript -e "mempalace::mempalace_cli()" ghcp doctor

The GHCP kit writes local resources under:

%USERPROFILE%\.copilot

It creates:

hooks\mempalace-ghcp.json
instructions\mempalace-memory.instructions.md
skills\ghcp-memory-recall
skills\ghcp-memory-save
mempalace\raw
mempalace\palace
mempalace\apps\r-shiny-memory-viewer

Standard CLI

# Initialize project room/entity files
Rscript -e "mempalace::mempalace_cli()" init C:\path\to\project

# Mine project files
Rscript -e "mempalace::mempalace_cli()" mine C:\path\to\project --wing my_project

# Mine conversation exports
Rscript -e "mempalace::mempalace_cli()" mine C:\path\to\convos --mode convos --wing chat_exports

# Search exact local memory
Rscript -e "mempalace::mempalace_cli()" search "why did we switch to GraphQL"

# Show recent wake-up context
Rscript -e "mempalace::mempalace_cli()" wake-up

Local Viewer

The GHCP installer writes a small R Shiny app to:

%USERPROFILE%\.copilot\mempalace\apps\r-shiny-memory-viewer\app.R

Run it with:

Rscript -e "shiny::runApp('$env:USERPROFILE\.copilot\mempalace\apps\r-shiny-memory-viewer\app.R', launch.browser = TRUE)"

Storage

The R fork stores drawers in local JSONL:

~\.mempalace\palace\drawers.jsonl

GHCP memory also keeps raw project/session ledgers under:

%USERPROFILE%\.copilot\mempalace\raw

Search is lexical/BM25-style plus exact-match boosting. See PORTING_NOTES.md for the differences from the Python fork.