memorax-ai/memorax-code

MemoraX Code – Persistent Context for AI Coding Assistants

What it is – An npm‑distributed tool that adds a shared, cloud‑backed memory layer to large‑language‑model coding agents such as GitHub Codex, Anthropic Claude Code, DeepSeek Harness, and OpenCode. It records and re‑uses engineering knowledge (lessons, architecture maps, preferences, procedures) across separate chat sessions so the assistant can pick up where it left off.

How it works

• Memory types – Four scoped stores:
  1. Coding Memory – reusable lessons, fixes, design rationales.
  2. Repo Memory – repository‑level facts (module ownership, entry points, recent PRs/commits).
  3. Personal Memory – user‑specific preferences (tone, language, output format).
  4. Procedure Memory – checklists and reusable steps for recurring tasks.
• The tool runs locally as a CLI (memorax-code) and talks to the MemoraX cloud service via an API key. It can also operate in a 90‑day guest mode without an account.
• When a coding agent finishes a turn, MemoraX Code extracts the salient instruction and the agent’s final response, then writes a concise entry to the appropriate memory store. It never uploads the full transcript.
• Agents can retrieve relevant memories on demand (e.g., /$memorax-code in Claude Code) or automatically when the system detects a matching context.

Key capabilities

Capability	What you get
Background memory write‑back	Lessons are saved automatically, no manual tagging needed
Preference continuity	Your preferred tone, language, and output style persist across sessions
Procedure reuse	Saved checklists are suggested for similar future tasks
Repo‑wide knowledge base	Architecture diagrams, entry points, and recent issue/PR data are kept up‑to‑date
Active memory control	CLI commands let you search, add, or delete entries manually
Multi‑client integration	Works with Codex, Claude Code, DeepSeek Harness, OpenCode; quota reminders are shown where supported

Getting started

1. Install globally – npm install -g @memorax/memorax-code (requires Node 20+, Python 3 for repo‑memory features).
2. Connect – Run memorax-code setup (creates a guest account) or memorax-code setup --existing-account after creating a MemoraX account and providing your API key.
3. Use – In a supported AI coding client, invoke the skill ($memorax-code or /memorax-code). The tool will automatically capture context and later retrieve it when you start a new session in the same repo.

Typical workflow

# Clone a repo you want the assistant to work on
git clone https://github.com/SWE-agent/test-repo.git && cd test-repo

# In the AI client, run the skill to build repo memory
$memorax-code   # (or /memorax-code in Claude Code)

# After a few turns, the assistant will have stored lessons.
# Close the chat, reopen a new session, and ask:
$memorax-code "Recall the engineering lesson and suggest checks"

The assistant will fetch the stored lesson and use it to guide the next task.

Management

• memorax-code status – view current configuration and cloud connection.
• memorax-code uninstall – clean up integrations while keeping local data (~/.memorax-code).
• All memories can be inspected, edited, or deleted in the MemoraX Console (https://platform.memorax.net/).

Security & privacy

• Only selected snippets (user instructions + final agent response) are sent to the cloud; raw model credentials and full traces stay local.
• API keys are stored in ~/.memorax-code/config.toml and must be kept private.
• Detailed network and retention policies are documented in SECURITY.md.

Who benefits

• Developers who rely on AI pair‑programmers and want the assistant to remember past fixes, architectural decisions, or personal coding style.
• Teams that use multiple AI coding agents across projects and need a consistent knowledge base.
• Anyone looking to reduce repetitive prompting and improve the efficiency of AI‑driven code reviews or feature implementation.

License – MIT.