إنتقل إلى المحتوى الرئيسي

Quick Start

Installing HMG​

HMG ships as a standalone binary via an official installer. The installer detects your platform, installs to the default directory, and configures PATH.

macOS / Linux​

bash
curl -fsSL https://github.com/HMG-AI/HMG-public/releases/latest/download/install.sh | sh

Windows​

powershell
irm https://github.com/HMG-AI/HMG-public/releases/latest/download/install.ps1 | iex

Verify the installation​

bash
hmg version
# Expected output, e.g.: hmg 1.7.8-developer

Upgrade​

bash

# Upgrade to the latest version (after installation)
hmg update

# To use a specific installer source
hmg update --installer-url <installer URL>

Uninstall​

Remove the install directory and clean up the PATH entry. Store data is not deleted automatically (your memories are kept); delete the store directory manually for a full cleanup.

Install failure? See Install failure.

Login and plans (optional)​

Basic memory features work without login. To unlock developer-edition features and a higher memory capacity:

  1. Upgrade your plan in the user center on the HMG website
  2. Log in locally:
bash
hmg login

Login requires network: HMG verifies your account with the remote service, identifies your plan, receives a signature, and unlocks the corresponding capabilities after local signature verification.

  • Login performs no data migration: local reads and writes always use your OS username as the tenant; the account is only used for identity mapping in cloud scenarios
  • Logging out or switching accounts only affects the cloud-side identity — local memories are untouched

Starting HMG locally​

After installation, let HMG prepare the local runtime. setup handles the daemon, the embedding model, and the components needed to run locally.

Prepare the runtime​

bash
# Preview what setup will do
hmg setup --dry-run

# Actually prepare the local runtime
hmg setup

Confirm it started​

bash
hmg doctor

doctor checks the core, store, integrations, and runtime state. On a fresh install it usually reports some agents as not integrated yet — that is normal.

Connect an agent​

If you want Codex, Cursor, or Claude Code to manage memory automatically in daily conversations:

bash
hmg init --agent codex --dry-run

Then run hmg init --agent <agent-id> for the agents you actually use. See Integration for details.

Write, recall, correct​

HMG memory operations come down to three core commands: write, recall, correct. Let's walk through a simple example.

1. Write a memory​

bash
hmg memorize "This project uses SQLite for the local cache; Redis is not introduced because we are offline-first" --source quick-start

A single self-contained sentence is enough: no decision: style prefix needed — HMG infers certainty, polarity, and other metadata automatically. The output returns an atom id (the memory's unique identifier) like atom-xxxx. Keep it — you'll need it in the next step.

2. Recall memories​

bash
hmg recall "what do we use for the local cache"

If the "SQLite cache" you just wrote shows up, the write-and-recall pipeline works.

3. Correct a memory​

When an old memory turns out wrong or outdated, correct it with correct instead of appending a new one:

bash
# Replace the content of the memory you just wrote
hmg correct atom-xxxx --action replace --reason "switched to LevelDB" --new-content "This project uses LevelDB for the local cache; Redis is not introduced because we are offline-first"

hmg recall "what do we use for the local cache" now returns the updated content.

Once an agent is integrated, the agent performs these three steps automatically at the right moments — no manual work needed. See Integration.

Next: Daily Usage Guide