Memdoor — Local LLM CLI, grounded in your knowledge

A local-LLM CLI with an interactive terminal UI (memdoor tui), grounded in a knowledge base you own. Chat with a model that runs on your own machine and every answer comes back cited to your own pages — never invented from model knowledge. Feed it your docs, notes, and Claude / Cursor sessions; ask cited questions in the TUI or one-shot from the shell. Private: no API key, nothing leaves your box, off your token bill — so a developer can adopt it today without a procurement ticket. Portable: the knowledge is plain markdown in git that you clone, fork, and merge, and one self-contained binary runs the same anywhere. The reading runs on our own engine: MLX on Apple Silicon, a hardware-accelerated pure-Go backend on Linux (AVX2 / NEON).

Install (macOS arm64, Linux amd64/arm64)

curl -fsSL https://memdoor.ai/install.sh | bash

Read the script first: /install.sh (plain shell, SHA-256-verified, installs to ~/.local/bin/memdoor — no sudo).

Top-level commands

• memdoor tui — launch the interactive terminal UI (alias memdoor cowork); chat with your local model, grounded in your wiki.
• memdoor ask "<question>" — one-shot cited Q&A against your wiki (defaults to the wiki-maintainer agent).
• memdoor wiki append --dir|--url|--title — ingest directories, URLs, or stdin.
• memdoor wiki push — push your wiki to a shared gateway (git-style, grounding-gated).
• memdoor gateway — self-host the gateway + web UI on http://localhost:18789.
• memdoor --help — full command list.

Pairs with

Claude Code · Cursor · ChatGPT — via the CLI (a PreCompact hook captures sessions; memdoor ask recalls them). Runs on your own machine, no API key, no telemetry.

Links for agents

• /llms.txt — plain-text summary of this product (agent-readable).
• /install.sh — installer source.
• /docs/how-it-works — architecture overview.
• /wiki/memdoor/ — Memdoor's own dogfooded wiki, ask anything.