G-Agent: Building AI Digital Roles with Persistent Identity and Visual Presence

Introduces G-Agent, an open-source runtime framework designed to build AI digital roles with persistent identity, long-term memory, visual presentation, multi-platform dialogue support, workflow automation, and self-hosted model routing. It aims to transform AI assistants from functional tools into continuous, personalized digital partners. Key keywords: AI role, digital assistant, long-term memory, multi-platform, workflow automation, self-hosted, privacy protection, open-source project.

Most current AI assistants are functional, transactional, and lack continuity/personality. G-Agent shifts to a "role-first" philosophy where memory, voice, visuals, and tools serve a unified identity instead of scattered functions. This allows AI to become a personalized digital partner present across multiple platforms (WhatsApp, Telegram, Discord, etc.) with consistent presence.

G-Agent is built on five core principles:

1. Role Consistency: Unified identity across memory, voice, visuals, and tools.
2. Presence Over Prompts: Integration into users' existing communication channels.
3. Utility Over Flash: Balances personality with practical tool integration and workflow execution.
4. Privacy Over Cloud Lock: Local memory, control, and clear permission policies.
5. Controllability Over Magic: Transparent AI behavior via allowlists, approval mechanisms, and limited tools.

G-Agent offers a comprehensive feature set:

• Role Creation & Visuals: Persistent identity with stable memory, unique tone, and visual rules (including situational selfies for visual presence).
• Multi-Platform Support: Seamless dialogue across WhatsApp, Telegram, Discord, Slack, email, and CLI with shared history and multi-modal replies.
• Google Workspace Integration: Limited access to Gmail, Calendar, Drive, Docs, Sheets, and Contacts via user authorization.
• Local Tools & Workflows: File operations, shell commands, scheduled tasks, media processing, and pre-built workflows (daily_brief, meeting_prep, etc.).
• Model Routing: Configurable via LiteLLM to connect to cloud APIs or self-hosted models (MiniMax, vLLM, etc.).
• Persistent Memory: Structured storage (MEMORY.md, PROFILE.md, PROJECTS.md, LESSONS.md) for continuous interaction.

G-Agent uses a modular architecture based on Python 3.11+ (Node.js 20+ for WhatsApp bridge). It supports Linux, macOS, Windows.

• One-Click Installation: Scripts for Arch, Debian/Ubuntu, macOS, and Windows PowerShell.
• Manual Installation: Clone the repo, install dependencies via pip, and configure using g-agent onboard.
• Optional Dependencies: espeak-ng (voice output), ffmpeg (media processing), GROQ_API_KEY (Whisper transcription).

G-Agent is suitable for various scenarios:

• Personal AI doppelgänger: Handle daily tasks (messages, emails, schedules) as a digital version of the user.
• Virtual Companion/Creative Partner: Use visual presence and memory for interactive companionship or creative collaboration.
• Workflow Automation: Automate tasks via Google Workspace and local tools (email sorting, document organization).
• Brand/Customer Service: Build brand-aligned AI roles for consistent multi-platform customer support.

G-Agent prioritizes privacy with a local-first architecture (sensitive data stays local). It uses clear permission systems to let users control access scope. This balances privacy protection with the flexibility of cloud service integration, reducing data leakage risks compared to fully cloud-based services.

G-Agent represents a shift from functional AI tools to personalized digital roles. It combines technical capability with role charm, local control with cloud integration, and privacy with utility. As AI evolves, digital roles will play a more important part in daily life. G-Agent's open-source nature allows community contributions, making it a promising platform for building custom AI roles.