Open Source AI Companions for Gaming: Super Agent Party, OpenJarvis & Project AIRI vs Questie (2026)

Why Open-Source AI Companions Are Having a Moment

Neuro-sama is the reference point for almost every project in this space. She's an AI VTuber who plays games, banters with chat, and has genuine personality — but she's closed-source and exists only during live streams. When the stream ends, she's gone. You can't talk to her, play games with her, or make her your own.

That gap triggered a wave of open-source projects from developers who wanted to own their own digital companion. The results are genuinely impressive. Project AIRI crossed 40,000 GitHub stars by mid-2026 — more than most commercial AI tools. Super Agent Party positions itself as "self-hosted Neuro-sama plus OpenClaw." OpenJarvis is backed by Stanford researchers and aims to be the PyTorch of local personal AI.

The movement is real and worth understanding. These are not half-finished weekend hacks.

Project AIRI (40.6k Stars): The Most Popular Open-Source AI Companion

Project AIRI (moeru-ai/airi) is the standout name in this space — the one most people mean when they say "self-hosted Neuro-sama." It runs on browsers via WebGPU and as a desktop app on Windows and macOS, supports VRM and Live2D avatars with auto-blink and eye tracking, integrates with Discord and Telegram, and has an actual Minecraft agent that plays autonomously. A Factorio integration is in progress.

The browser version at airi.moeru.ai is the fastest way to try it — no install required. The desktop version needs pnpm and a Node runtime. The Minecraft gameplay feature needs a Minecraft server running and the agent configured separately. The project's own README states clearly: "We are still in the early stage of development where we are seeking out talented developers to join us." It's built by 160 contributors across 3,800+ commits. Updates ship fast, but the contributor who maintains a specific feature might step away — that's the reality of community-maintained software.

If you want to play Minecraft with an open-source AI companion that actually builds and explores autonomously, AIRI is the most serious attempt anyone has made. If you want a voice companion for your CS2 session tonight without configuring anything, that's a different answer.

Super Agent Party (SAP, 2.3k Stars): "Self-Hosted Neuro-sama + OpenClaw"

Super Agent Party describes itself directly: "An AI desktop companion with endless possibilities. Super Agent Party = Self-hosted Neuro-sama + OpenClaw." The scope is wide. SAP gives you a VRM desktop companion with customizable animations and voice interaction, a computer control system where the AI can play games autonomously using desktop vision, multi-platform bot deployment across Discord, Telegram, WeChat, Feishu, Slack, and live streaming to Bilibili, YouTube, and Twitch.

It's also where scope becomes a liability for casual users. The Windows install is relatively smooth — download, extract, run the bat file. macOS requires removing quarantine attributes via terminal, a step that trips up plenty of non-developers. Docker deployment means the animated companion only shows in a browser window. The extension ecosystem is impressive (AI galgame, AI tarot, Live 2D, RSS reader) but each extension is its own install. And the disclaimer section of the README is seven bullet points long, covering instability of third-party APIs, legal use requirements, and the developer's limits of responsibility.

SAP is AGPL-licensed, which means any commercial fork must also be open source. The project is actively maintained with 33 releases as of mid-2026 and a 5,698-commit history. For technically capable users who want to build a local AI ecosystem with genuine computer automation, SAP is genuinely capable. For someone who wants to game with an AI companion and not think about infrastructure, the learning curve is steep.

OpenJarvis (5.9k Stars): Stanford's Local-First Personal AI

OpenJarvis is different from AIRI and SAP in one important way: it's a research project with serious institutional backing. It comes out of Hazy Research and the Scaling Intelligence Lab at Stanford SAIL, and its paper is on arXiv. The framing is also different — OpenJarvis wants to be the foundational framework for local AI, the "PyTorch of personal AI," not a companion app.

It has eight built-in agents: a morning digest that reads email and calendar, a deep research agent, a code assistant, a scheduled monitor, and a simple chat agent. Install is genuinely fast — a one-liner curl command handles the Python virtual environment, Ollama, and a starter model in about three minutes. The jarvis CLI then launches the experience.

OpenJarvis is not a gaming companion. Its sponsors include Google Cloud Platform, Lambda Labs, and IBM Research — it's built to run research workloads and productivity agents locally. There's no real-time game spectating, no voice chat tuned for active gameplay, no companion personality built around gaming culture. The skills system (importing from Hermes Agent, OpenClaw skills) is interesting for power users who want to build custom agent behaviors.

If you are a developer interested in local AI research or building agents that run on-device, OpenJarvis is worth exploring. For gaming, it's the wrong tool.

Others Worth Knowing in 2026

The open-source AI companion ecosystem extends beyond these three. A few projects worth knowing:

• N.E.K.O. — A Chinese-developed companion with five-dimensional memory (working, recent, factual, reflective, personality), Live2D/VRM/MMD avatars, and a Steam page. It includes computer control via CUA and OpenClaw agent-to-agent protocols. Community is primarily Chinese-speaking.
• z-waif — Described by AIRI's own README as "great at gaming, autonomous, and prompt engineering." Focused on autonomous gameplay and deeper prompt engineering customization. Requires Python setup.
• Open-LLM-VTuber — A streaming-focused VTuber framework with Live2D and VRM support. Well-suited for content creators who want to run an AI VTuber without cloud dependency. Setup is non-trivial.
• kimjammer/Neuro — A Neuro-sama recreation originally built in seven days. Historically notable, and AIRI cites it as a well-completed implementation. Less actively maintained than the others.

All share the same core trade-off: powerful, customizable, yours to own — but requiring real investment to get running and keep running.

What These Projects Have in Common: The Setup Wall

Every open-source AI companion project hits the same pattern. The README is impressive. The demo video is impressive. Then you read the installation section.

Project AIRI: pnpm i then pnpm dev for the browser version — straightforward if you have Node and pnpm. The desktop version needs additional steps, the Minecraft agent runs separately, and the Nix build adds another layer for Linux users.

Super Agent Party on macOS: download, extract, run sudo xattr -rd com.apple.quarantine [folder] in terminal, then chmod +x the startup scripts, then run them. The README explicitly notes: "The operating system must be Windows 10/11, Windows Server 2025, or a later version" for the easy installer.

OpenJarvis: the curl one-liner is genuinely easy, but then you're in a CLI running jarvis init --preset morning-digest-mac, connecting Google Drive via OAuth, managing API keys, and reading preset documentation before your first meaningful session.

These are not flaws in the projects — they're inherent to the self-hosted software model. If you want to own the software, you have to run it. That's a legitimate trade-off. It just means the audience is developers, homelab enthusiasts, and technically capable users — not gamers who want an AI companion before their next session.

Where Questie Fits Into This Picture

Questie occupies the opposite end of the spectrum. It's a maintained, gaming-first AI companion platform — no self-hosting, no CLI, no configuration. Sign up, pick a companion, and you have voice chat and screen vision during your next gaming session.

The voice chat runs on LiveKit at sub-300ms latency — tuned specifically for the interaction pattern of active gaming, where you're talking to an AI while also paying attention to a game. The screen vision captures your gameplay in real-time and lets the AI react to what's actually happening — your health bar, the boss pattern, the clutch play — without you having to describe anything. Persistent memory carries context across sessions, so the companion remembers your playstyle, past conversations, and preferences.

There are 25+ pre-built companion personalities designed around gaming. You can also create your own companion without touching a config file — just a form with name, personality, and voice. For Twitch streamers, there's a built-in co-host mode that integrates with your stream.

The honest trade-off: Questie is cloud-hosted, which means you don't own the infrastructure. Your data lives on Questie's servers. There's no VRM avatar on your desktop. A desktop app is in active development but not yet shipped. For users who specifically want local control and a visible avatar companion, the open-source projects above have things Questie currently doesn't. That's worth saying directly.

What Open-Source AI Companions Do Genuinely Well

A fair comparison has to acknowledge where these projects beat Questie outright.

• Data ownership. Everything runs on your hardware. Your conversations, your companion memory, your voice data — none of it passes through a third-party server. For users who care deeply about privacy, self-hosting is the only real answer.
• VRM and Live2D avatars. Super Agent Party and AIRI both render animated 3D companions on your desktop with blinking, idle animations, and lip sync. This is the closest thing to having Neuro-sama as your own desktop companion. Questie doesn't have this yet (desktop app in development).
• Deep customization. You can fork the code, change anything, add integrations, write new skills. The ceiling is the LLM itself. Questie's customization is limited to what's exposed through the UI.
• Cost at scale. Once you have your own API keys set up, heavy usage doesn't scale with a fixed monthly subscription. If you're running an AI companion for eight hours a day, the economics of self-hosting can tip in your favor depending on model choice.
• Community and creativity. The open-source communities around these projects are active and creative. New extensions, plugins, and use cases appear weekly. If you want to be part of building AI companion software, these are the projects where your contributions matter.

Who Should Use Which?

Use an open-source AI companion if:

• You're comfortable with a terminal and don't mind configuring environments
• Data ownership and local-first operation are non-negotiable for you
• You want a VRM avatar on your desktop right now
• You want to contribute to the project and shape its direction
• You play Minecraft specifically and want AIRI's autonomous gameplay agent
• You run many long daily sessions and want control over LLM inference costs

Use Questie if:

• You want an AI companion in your gaming session without any setup
• Real-time voice chat while you play is the core feature you need
• You want the AI to watch your screen and react to gameplay — without configuring a computer vision pipeline
• You stream on Twitch and want a co-host that works tonight, not next weekend
• Predictable monthly pricing matters more than paying per token
• You want a maintained platform with regular improvements and a dedicated support team

The Honest Bottom Line

The open-source AI companion space in 2026 is producing genuinely impressive software. Project AIRI's 40,000+ stars and 78 releases are not a fluke — the Minecraft gameplay agent alone is more technically ambitious than most commercial products have attempted. Super Agent Party's computer control, live streaming, and multi-platform bot deployment put it ahead of many SaaS tools in scope. OpenJarvis's Stanford backing and research-grade evaluation framework are real.

But all three share a common honest limitation: they require you to be a builder to use them well. And building a local AI companion setup is a project in its own right — one that takes days to get right and ongoing maintenance to keep running.

Questie exists for the gamer who wants the companion experience, not the builder experience. The setup gap is real. The gaming-specific voice and screen features are real. The fact that you can try it tonight, for free, without reading a README — that's the value.

Use both if you can. Self-hosted AI companions and Questie are not really competing — they're serving different moments for different people. But if you have a raid tonight and you want someone in your ear who's actually watching, Questie is the answer right now.

Frequently Asked Questions

What is Super Agent Party (SAP)?

Super Agent Party (heshengtao/super-agent-party) is an open-source, self-hosted AI companion platform that bills itself as "self-hosted Neuro-sama + OpenClaw." It combines a VRM 3D desktop companion, multi-platform instant messaging bot deployment (Discord, Telegram, WeChat, Slack), live streaming to YouTube, Twitch, and Bilibili, computer control via desktop vision, and SillyTavern character card support for roleplay. It's AGPL-licensed with 2,300+ GitHub stars and 33 releases. Windows and macOS installs are available; Linux has AppImage and deb packages. Best suited for technically capable users who want a deeply featured local AI ecosystem rather than a simple gaming companion.

What is OpenJarvis?

OpenJarvis (open-jarvis/OpenJarvis) is a Stanford-backed open-source framework for local-first personal AI. It runs AI agents using Ollama with eight built-in agents including daily briefings, deep research, code assistance, and persistent monitoring. It has 5,900+ GitHub stars and installs via a curl one-liner in about 3 minutes on broadband. OpenJarvis is designed as a research-grade productivity platform, not a gaming companion. It has no real-time game spectating or voice chat for gaming, and is sponsored by Google Cloud Platform, Lambda Labs, and IBM Research.

What is Project AIRI?

Project AIRI (moeru-ai/airi) is an open-source recreation of Neuro-sama — the famous AI VTuber. With 40,600+ GitHub stars, it's the most popular open-source AI companion project as of mid-2026. AIRI supports VRM and Live2D avatars, real-time voice chat, Minecraft and Factorio gameplay, Discord and Telegram integration, and local WebGPU inference. It runs in browsers and as a desktop app on Windows and macOS. The project is MIT-licensed and under active development with 78 releases. Its own README notes it's in early development and actively seeking contributors.

Are there apps like Super Agent Party for gaming?

The closest pre-built alternative specifically for gaming is Questie — real-time voice chat during gameplay, live screen vision, and gaming companion personalities without any technical setup. Other apps in the open-source space include Project AIRI (game-playing AI VTuber), z-waif (autonomous gaming assistant), Open-LLM-VTuber (streaming-focused), and N.E.K.O. (memory-focused companion with Steam page). All of the open-source options require technical setup; Questie is the ready-to-use option.

Is Questie a good alternative to Project AIRI?

For gaming sessions where you want an AI companion without setup: yes. Questie has real-time voice chat and screen vision ready to go. Project AIRI has things Questie doesn't — a VRM desktop avatar, autonomous Minecraft gameplay, and full local operation. If those specific features matter to you and you're comfortable with a pnpm dev environment, AIRI is worth exploring. If you want a companion for your gaming session tonight with no setup, Questie is the answer.

What is the difference between self-hosted and cloud-hosted AI companions?

Self-hosted AI companions (AIRI, SAP, OpenJarvis) run on your own hardware — your data stays local, you control the LLM, and you pay your own API costs or run models offline. Cloud-hosted companions like Questie run on maintained servers — setup is instant, features are managed for you, and pricing is predictable, but your data passes through the platform. Both models have genuine merit. The choice depends on how much you value data ownership versus convenience.

Can I use both an open-source AI companion and Questie?

Yes — they serve different contexts well. Open-source companions are strong when you want a local ecosystem, a VRM avatar on your desktop, or deep customization. Questie is strong when you want a companion in your active gaming session with minimal friction. Some users run Super Agent Party for general AI interaction during the day and switch to Questie when they sit down for a gaming session. The use cases don't overlap significantly.