memCode

memCode is a terminal coding agent focused on starting quickly, editing code, and working with OpenRouter, GPT/OpenAI-compatible APIs, or Google Gemini.

XMem memory is optional. You can connect it later if you want persistent project memory.

Install

npm install -g @xmem-ai/memcode

For beta testing, use the npm package. Source installs are not supported from a public repository yet.

Configure A Provider

Use at least one provider key:

# OpenRouter
export OPENROUTER_API_KEY=sk-or-v1-...

# GPT / OpenAI
export OPENAI_API_KEY=sk-...

# Any OpenAI-compatible /v1/chat/completions endpoint
export XCODE_API_URL=http://127.0.0.1:8000/v1
export OPENAI_API_KEY=local-or-real-key

# Google Gemini
export GEMINI_API_KEY=...

Run

memcode

Non-interactive mode:

memcode -p --bare --model gpt "explain this repo"

Models

# GPT / OpenAI-compatible
memcode --model gpt
memcode --model gpt-4o
memcode --model gpt-4o-mini
memcode --model gpt-5-mini
memcode --model gpt-5.4-mini
memcode --model gpt-5.4-nano

# Gemini direct when GEMINI_API_KEY is set
memcode --model gemini
memcode --model gemini-2.5-pro

# OpenRouter
memcode --model kimi
memcode --model deepseek
memcode --model llama4

Inside memCode, use /model to switch models and inspect cost notes.

Optional XMem

If you run an XMem server, point memCode at it:

export XMEM_API_URL=http://localhost:8000
export XMEM_API_KEY=...

Without XMEM_API_URL, memCode runs without XMem network calls.

Useful Commands

| Command | Description | | --- | --- | | /model or /model <name> | Switch model and view costs | | /compact | Compact conversation context | | /clear | Clear conversation history | | /quit | Exit memCode |

Development

npm run build

The build checks the CLI entrypoint, non-interactive doctor output, provider routing smoke tests, and npm package contents.