@remnic/server

Standalone Remnic memory and context server -- HTTP and MCP interfaces without requiring OpenClaw.

Part of Remnic, open-source memory and context for user-aware agents.

Install

npm install @remnic/server

Most users should install @remnic/cli instead, which includes the server and provides daemon management via remnic daemon start.

What it does

The server exposes Remnic's memory engine over two interfaces:

• HTTP API -- RESTful endpoints for memory recall, observe, search, and management
• MCP (Model Context Protocol) -- tool-based interface for AI agents that support MCP (Replit, Cursor, etc.)

Both interfaces connect to the same @remnic/core engine. All data stays local on your filesystem.

Usage

Run the standalone server:

npx --package @remnic/server remnic-server --help
npx --package @remnic/server remnic-server --port 4318

Run the container image:

docker run --rm \
  -p 4318:4318 \
  -v remnic-data:/data \
  -e REMNIC_AUTH_TOKEN=change-me \
  ghcr.io/joshuaswarren/remnic:latest

The image stores memory and runtime configuration under /data. The container enables the admin console explicitly and serves it at /engram/ui/. Package installs keep the console disabled unless server.adminConsoleEnabled or REMNIC_ADMIN_CONSOLE_ENABLED=true is set with a public asset directory.

For the public ChatGPT developer-mode flow (Tailscale Funnel / Cloudflare Tunnel + OAuth 2.1 with local operator approval), see docs/integration/chatgpt.md.

The package also ships the legacy engram-server binary for compatibility. The bin wrappers are source-controlled so package managers can link them during workspace installs; release builds verify that both targets have Node shebangs and can start their help command before publish.

import { startServer } from "@remnic/server";

const server = await startServer({
  port: 3141,
  authToken: process.env.REMNIC_AUTH_TOKEN,
});

console.log(`Remnic server listening on http://${server.host}:${server.port}`);
await server.stop();

License

MIT