Open source AI girlfriend github — how it works
By FlirtAlly · October 8, 2026
Open source AI girlfriend github — how it works
Key Facts
- Airi, the most popular open-source AI companion on GitHub, has 50.1k stars and even plays Minecraft and Factorio per its project page.
- MyGirlGPT runs four services on three ports — LLM on 5001, TTS on 6006, and Stable Diffusion on 7860 — according to its documentation.
- The DIY media stack leans on mature tooling: stable-diffusion-webui (157k stars), ComfyUI (89.8k), and Bark TTS (38.5k) ecosystem data shows.
- "Open source" often means open wrapper code, not open models — Aion runs on Anthropic's proprietary Claude API per its project page.
- MyGirlGPT lists long-term memory, video messages, and personality switching as roadmap items, not shipped features per its README.
- The generic GitHub "girlfriend" topic is dominated by non-AI projects — only SoulSync.AI_Waifu (8 stars) is AI-focused topic listings show.
- MyGirlGPT's LLM is the conductor: it autonomously decides when to trigger Stable Diffusion image generation during conversation per its architecture docs.
The Open-Source AI Girlfriend Landscape on GitHub
The open-source AI girlfriend landscape on GitHub reveals a niche market shaped by technical experimentation and privacy-focused ideals. While the "girlfriend" topic hosts mostly non-AI projects, two exceptions stand out: Airi with 50.1k stars and MyGirlGPT, a modular system leveraging local LLMs and media pipelines. These projects highlight a DIY ethos centered on self-hosting, privacy, and full control over AI interactions.
Most open-source AI girlfriend projects prioritize local media generation through tools like Stable Diffusion for selfies, Bark TTS for voice, and Live2D/VRM avatars. MyGirlGPT, for instance, runs on personal servers with a Telegram bot interface and a Pygmalion-Vicuna-7b LLM, while Airi offers real-time voice chat and Minecraft integration. However, features like long-term memory and video messaging remain underdeveloped, with roadmap notes rather than implemented capabilities.
- Self-hosting requires technical setup, including server configuration and port management
- Media tools like Stable Diffusion WebUI (157k stars) and Bark (38.5k stars) form the backbone of DIY projects
- Open-source claims often apply to code, not models—Aion uses Anthropic’s Claude, while MyGirlGPT relies on local LLMs
The ecosystem reflects a trade-off between flexibility and accessibility. While projects like Airi and MyGirlGPT offer full control over data and identity, they demand significant user effort. FlirtAlly contrasts this approach by delivering identity-locked media (selfies, voice notes) with unlimited memory across all plans, eliminating setup barriers. For users seeking privacy without technical overhead, this represents a distinct value proposition.
< strong class="blog-highlight">Self-hosting and full control remain central to the open-source appeal, but the complexity often limits adoption. FlirtAlly’s model—prioritizing ease of use while retaining privacy—addresses this gap, positioning itself as an alternative for those who want AI companions without the DIY burden.
Join the waitlist — be first in to experience a different approach to AI companionship.
How These Projects Work: A Modular Stack
Under the hood, most open-source AI girlfriend projects on GitHub aren't a single program — they're a small orchestra of services, with a large language model acting as the conductor. Understanding that architecture explains both what these projects can do and where they fall short.
The clearest documented example is MyGirlGPT, which runs four communicating services on a personal server: a Telegram bot for messaging, an LLM server on port 5001, a TTS server on port 6006, and a Stable Diffusion WebUI instance on port 7860 for generating selfies (project documentation). The LLM is the brain — it doesn't just write replies, it "decides when to trigger image generation" and calls the image server's API on its own.
The stack looks like this:
- A messaging layer (usually Telegram) that relays text between you and the model
- A local LLM — MyGirlGPT uses TehVenom/Pygmalion-Vicuna-1.1-7b — for conversation and decision-making
- A TTS engine like Bark that turns replies into voice messages
- An image pipeline, typically Stable Diffusion WebUI, for photorealistic selfies on request
Other projects swap components in and out. Airi, the most popular companion project at 50.1k stars, replaces the selfie pipeline with Live2D and VRM avatars and realtime voice chat (GitHub's ai-companion topic). Aion pushes further into 3D with an R3F body built on React Three Fiber (companion-ai topic). The underlying tooling is mature — stable-diffusion-webui holds 157.0k stars, ComfyUI 89.8k, and Bark 38.5k (ecosystem data).
One nuance matters: "open source" often means open wrapper code, not open models. Aion runs on Anthropic's proprietary Claude API, and SoulSync.AI_Waifu calls Gemini — only the glue code is public (topic listings). MyGirlGPT is more genuinely self-contained, but even it lists long-term memory and video messages as roadmap items, not shipped features (README).
That's the honest trade-off of the DIY route: you get full control and privacy, but you assemble identity, voice, and memory yourself. Managed platforms move in the opposite direction — FlirtAlly, for instance, ships at launch with identity-locked companions whose every photo and video is checked against a canonical face, plus memory on every plan, no server ports required. The modular stack is powerful; it just asks you to be the engineer.
