Galyarder Agent

Agentic Intelligence characters for real workflow, can send selfie or mirror with consistency face in ur agents, also have long-term memory about u, automation ur work, and self-hosted model routing.

What Is g-agent?

g-agent is an open-source assistant runtime you run on your own machine, with your own policies, through channels you already use (CLI, Telegram, WhatsApp).

It is built for one outcome: practical automation without losing control.

Why This Project Exists

Most assistant projects drift to one of two extremes:

• feature-heavy platforms with unclear internals,
• minimal demos that look clean but break in real operations.

g-agent is built in the middle.

It keeps the powerful parts (agent loop, tools, memory, scheduling, integrations) while keeping the runtime understandable and auditable.

If your assistant can act but you cannot explain what it can access, who can talk to it, and where it runs, it is not really your assistant.

Philosophy

• Useful over flashy: solve real tasks first.
• Understandable over abstract: keep code and behavior inspectable.
• Private over cloud-lock: local memory, local control, explicit policy.
• Controlled over magical: allowlists, approvals, scoped tools.

If your assistant cannot run reliably on your own machine, it is not your assistant.

What You Can Do Today

• Chat through CLI, Telegram, or WhatsApp with persistent history and enhanced editing.
• Run models through configurable OpenAI-compatible proxies (MiniMax, CLIProxyAPI, vLLM, LiteLLM, etc.) via LiteLLM routing.
• Keep durable memory across sessions (MEMORY.md, PROFILE.md, PROJECTS.md, LESSONS.md).
• Schedule recurring jobs and proactive reminders.
• Run workflow packs like daily_brief, meeting_prep, and inbox_zero_batch.
• Send multimodal replies (text, image, voice, sticker, document).
• Connect Google Workspace (Gmail, Calendar, Drive, Docs, Sheets, Contacts).

Requirements

• Linux, macOS, or Windows
• Python 3.11+
• Node.js 20+ (for WhatsApp bridge)
• Optional: espeak-ng + ffmpeg (generated voice-note output)
• Optional: GROQ_API_KEY (voice-note transcription via Whisper)

Quick Start

Option A: Installer script (recommended)

Option B: From source

git clone https://github.com/galyarderlabs/galyarder-agent.git
cd galyarder-agent/backend/agent
pip install -e .

First run

g-agent onboard
g-agent status
g-agent gateway

Then set provider/model in ~/.g-agent/config.json.

Installer flags (optional)

• Common:
  • G_AGENT_INSTALL_DIR=/path/to/repo (default: ~/galyarder-agent)
  • G_AGENT_DATA_DIR=/path/to/data (default: ~/.g-agent)
• Arch:
  • G_AGENT_SKIP_PACMAN=1
  • G_AGENT_SKIP_SERVICES=1
  • G_AGENT_AUTO_START_SERVICES=0
• Debian/Ubuntu:
  • G_AGENT_SKIP_APT=1
  • G_AGENT_SKIP_SERVICES=1
  • G_AGENT_AUTO_START_SERVICES=0
• macOS:
  • G_AGENT_SKIP_BREW=1
  • G_AGENT_SETUP_LAUNCHD=1
  • G_AGENT_AUTO_START_SERVICES=0
• Windows:
  • G_AGENT_SKIP_WINGET=1
  • G_AGENT_SETUP_TASKS=1

Uninstall

Optional flags:

• G_AGENT_REMOVE_SERVICES=0 keep startup services/tasks
• G_AGENT_REMOVE_REPO=1 remove repo directory
• G_AGENT_WIPE_DATA=1 remove full ~/.g-agent data

Usage Examples

From CLI:

g-agent agent -m "Summarize my priorities for today."
g-agent agent -m "/pack daily_brief focus revenue --sticker --silent"
g-agent proactive-enable

From Telegram/WhatsApp:

• Ask questions.
• Request reminders.
• Trigger workflow packs.
• Send images/voice and get multimodal responses.

Core Commands

Channel Setup

Supported channels and typical setup effort:

* Experimental in current release.

Telegram

{
  "channels": {
    "telegram": {
      "enabled": true,
      "token": "BOTFATHER_TOKEN",
      "allowFrom": ["123456789"]
    }
  }
}

WhatsApp

{
  "channels": {
    "whatsapp": {
      "enabled": true,
      "bridgeUrl": "ws://localhost:3001",
      "allowFrom": ["6281234567890"]
    }
  }
}

Pair WhatsApp:

g-agent channels login
g-agent gateway

Discord / Feishu (experimental)

• See docs/channels.md for full step-by-step setup.
• Keep allowFrom strict for any public-facing deployment.

Google Workspace (OAuth)

g-agent google configure --client-id "YOUR_CLIENT_ID" --client-secret "YOUR_CLIENT_SECRET" --calendar-id "primary"
g-agent google auth-url
# open URL, approve consent, copy value after ?code=
g-agent google exchange --code "PASTE_CODE"
g-agent google verify

Default scopes include:

• gmail.modify
• calendar
• drive.readonly
• documents
• spreadsheets
• contacts.readonly

Memory Model

Memory lives in workspace/memory:

• MEMORY.md: long-term notes
• PROFILE.md: identity/preferences
• RELATIONSHIPS.md: people context
• PROJECTS.md: active project context
• LESSONS.md: behavior improvements
• YYYY-MM-DD.md: daily memory notes

Architecture

Execution flow:

Channel Input -> Message Bus -> Agent Loop -> Tools + Memory + Scheduler -> Outbound Dispatcher

Runtime model:

• Python process for agent runtime
• Node.js bridge for WhatsApp transport
• local filesystem for state and memory

Security Model (Plain Language)

• tools.restrictToWorkspace keeps file/shell access inside allowed workspace.
• channels.*.allowFrom controls who can send messages to the assistant.
• approval mode can require confirmation for risky actions.
• policy presets (personal_full, guest_limited, guest_readonly) control tool risk.
• separate profiles via G_AGENT_DATA_DIR isolate personal and guest environments.

For details, read SECURITY.md and backend/agent/SECURITY.md.

24/7 Service Mode (systemd --user)

systemctl --user enable --now g-agent-wa-bridge.service
systemctl --user enable --now g-agent-gateway.service

Check:

systemctl --user status g-agent-wa-bridge.service
systemctl --user status g-agent-gateway.service

Optional lingering:

sudo loginctl enable-linger "$USER"

Docker Quick Run

docker build -t g-agent ./backend/agent
docker run -v ~/.g-agent:/root/.g-agent --rm g-agent g-agent onboard
docker run -v ~/.g-agent:/root/.g-agent --rm g-agent g-agent status
docker run -v ~/.g-agent:/root/.g-agent -p 18790:18790 g-agent g-agent gateway

Guest Profile Isolation

mkdir -p ~/.g-agent-guest
G_AGENT_DATA_DIR=~/.g-agent-guest g-agent onboard
G_AGENT_DATA_DIR=~/.g-agent-guest g-agent status

Each profile has isolated config, memory, cron jobs, bridge data, and OAuth/session artifacts.

Troubleshooting

Telegram timeout

curl -sS "https://api.telegram.org/bot<YOUR_TOKEN>/getMe"

WhatsApp bridge reconnect loops

g-agent channels login --rebuild
systemctl --user restart g-agent-wa-bridge.service g-agent-gateway.service

Production Checklist

1) Lock access

• Keep channels.*.allowFrom non-empty on enabled channels.
• Keep tools.restrictToWorkspace: true.
• Keep tools.approvalMode: "confirm" or stricter.
• Use separate G_AGENT_DATA_DIR for guest/public assistants.

2) Monitor health

g-agent doctor --network
systemctl --user status g-agent-gateway.service g-agent-wa-bridge.service
journalctl --user -u g-agent-gateway.service -u g-agent-wa-bridge.service -n 120 --no-pager

3) Backup critical state

mkdir -p ~/.g-agent-backups
tar -czf ~/.g-agent-backups/g-agent-$(date +%F).tar.gz \
  ~/.g-agent/config.json \
  ~/.g-agent/workspace/memory \
  ~/.g-agent/cron

4) Rotate keys safely

NEW_TG_TOKEN='YOUR_NEW_TOKEN'
tmp=$(mktemp) && jq --arg v "$NEW_TG_TOKEN" '.channels.telegram.token = $v' ~/.g-agent/config.json > "$tmp" && mv "$tmp" ~/.g-agent/config.json
systemctl --user restart g-agent-gateway.service

FAQ

Why choose g-agent?

Choose g-agent if you want a lean runtime with faster auditability and simpler day-to-day operations.

What makes g-agent practical for daily use?

It emphasizes focused workflows, policy presets, workflow packs, and operational docs that stay close to real runtime usage.

Is this suitable for production personal use?

Yes, with proper allowlists, workspace restrictions, approval settings, and credential hygiene.

Can I use this as my always-on personal assistant?

Yes. The typical production flow is user services + allowlists + restricted workspace + approval mode.

Is it safe enough for personal data?

It is built with practical controls (allowFrom, restrictToWorkspace, approvals, profile isolation).
You still need to review your configuration and keep tokens scoped/rotated.

Runtime Roadmap

Roadmap and implementation status:

• docs/roadmap/runtime-roadmap.md

Documentation

• Docs site: https://galyarderlabs.github.io/galyarder-agent/
• Getting started: docs/getting-started.md
• Configuration: docs/configuration.md
• Channels: docs/channels.md
• Install matrix: docs/install-matrix.md
• Operations: docs/operations.md
• Troubleshooting: docs/troubleshooting.md
• FAQ: docs/faq.md
• Backend docs: backend/agent/README.md
• Roadmap: docs/roadmap/runtime-roadmap.md
• Changelog: CHANGELOG.md

Contributing

We welcome focused contributions:

• security hardening,
• reliability improvements,
• performance improvements,
• documentation clarity.

Please read CONTRIBUTING.md and CODE_OF_CONDUCT.md before opening a PR.

License

MIT — see LICENSE.

Star History

“Digital sovereignty is not isolation — it is ascendancy with ownership: your memory, your tools, your systems, your future.”