Ram0

Ram0 is an unofficial, self-hosted fork derived from Supermemory. It adds an authenticated HTTP/MCP gateway, a self-hosted graph UI, Claude Code and Codex integrations, guarded Unraid deployment, and resumable Mem0 migration tooling.

Ram0 is independently maintained and is not sponsored, endorsed, or supported by Supermemory.

What this fork adds

• Streamable HTTP MCP tools for explicit recall, capture, listing, and inspection.
• Automatic recall and capture hooks for Claude Code and Codex.
• Stable repository-scoped memory shared across clones and linked worktrees.
• A bearer-authenticated compatibility gateway for the local Supermemory engine.
• A graph interface that reads the self-hosted API rather than the hosted service.
• Resumable, count-verified migration from two Ram0/Mem0 user accounts.
• A guarded Unraid cutover that preserves the legacy database and ports.

Architecture

Claude Code / Codex -> HTTPS gateway -> local Supermemory engine

The gateway also exposes /mcp; the graph application uses the same gateway. The default deployment binds the API to port 18888 and the graph application to port 13000. Credentials are provided at runtime and never belong in Git.

Deploy

Read deploy/supermemory/README.md. The deployment script stages candidate containers, verifies them, backs up the legacy database, and promotes only after readiness checks pass.

Install Claude Code and Codex integrations

Requirements: Node.js 20+, current Claude Code, current Codex, and the gateway key in SUPERMEMORY_API_KEY.

node integrations/coding-agents/install.mjs \
  --base-url https://brain-api.example.com \
  --force

source ~/.config/ram0-supermemory/env.sh

Keep the key in a mode-0600 secret file or secret manager and source it before the generated helper. Never place it in this repository or a command argument. See integrations/coding-agents/README.md for installation, project scoping, migration from the old plugin, and checks.

Migrate Mem0 memories

The importer is resumable and verifies source-to-destination counts. Read tools/ram0-migration/README.md before running it. Preserve the source database and export until the destination audit passes.

Upstream maintenance

origin is the Ram0 fork and upstream is https://github.com/supermemoryai/supermemory.git. Upstream changes are brought in through a dedicated sync branch, followed by tests, license review, and a reviewed pull request. Ram0 release tags are immutable.

Security and distribution

• The gateway requires bearer authentication for API and MCP requests.
• Runtime secrets and migration exports are excluded from Git.
• The legacy database is retained as a rollback source.
• The engine build downloads an official, version- and checksum-pinned supermemory-server release asset.
• The release asset does not currently state sufficiently explicit binary redistribution terms. Images containing it must remain private and the binary must not be attached to a public Ram0 release unless Supermemory supplies written license clarification.

License and attribution

The upstream source is distributed under the root MIT license, whose copyright and permission notice are retained unchanged. Ram0 additions are also MIT licensed. skills/supermemory is a separately licensed Apache-2.0 component and retains its own license.

See NOTICE for provenance and docs/research/2026-09-12-supermemory-fork-licensing.md for the technical licensing review. Ram0 is not legal advice and does not claim rights that an upstream dependency or release asset has not documented.