⚡ Run it on your own computer

Generations run on one shared GPU, so at busy times you wait in a queue. If that's too slow, you can run your tier's AI model locally on your own machine - instant, private, and no queue. It still counts toward your plan's weekly limit, so it stays fair. This page checks what your computer can handle.

🚀 The easy way - one download sets it all up

Don't want to touch a terminal? Download AGenO Local - a little app that installs ComfyUI + the models, starts everything with the right settings, and gives you a dashboard with Install / Start buttons. You're signed in automatically.

Then run it: python ageno-local.py (Windows: double-click, or py ageno-local.py). A dashboard opens in your browser → click Install everything, then Start. Prefer to do it by hand? The exact commands are still below.
🔒 Is it safe to run? What this app can and can't do
Fair question - you should always be careful about running downloaded software. Here's the honest picture:
  • It's plain-text Python you can read first. ageno-local.py isn't a compiled black box - open it in any text editor before running and you can see exactly what it does. Every command is right there.
  • What it does: installs Ollama + ComfyUI and the AI models, starts them on your computer, and opens a local dashboard. It signs you in by including your own AGenO token so local generations count against your plan.
  • What it talks to: the AI services running on your machine (Ollama, ComfyUI) and agenoai.com (only to check your plan/quota). It does not scan, upload, or read your personal files - it only installs the model software into its own folder.
  • Honest limitation: this is a solo project and the installer has not had a formal third-party security audit yet, and I don't publish signed releases or checksums today. If that matters for your situation, run it on a spare/personal machine rather than a work computer with sensitive data - or just skip it and use the cloud version, which needs no install at all.
Found a security issue? Email [email protected] - I'll respond fast.

1 · Check your hardware

- or set your GPU memory and system RAM below -
Why both? A model runs fastest fully in VRAM. If it doesn't fit, it spills into your system RAM (still works, just slower). With little of either, it won't run.
Not sure? Windows: Ctrl+Shift+Esc → Performance (GPU + Memory). Mac: Apple menu → About This Mac.

2 · What you can run

Each AGenO tier maps to one model - to use AGenO's models locally you run your tier's model, not a random one. ✅ runs great · ⚠️ runs but slow · ❌ won't fit.

Your tierModelNeeds ~VRAMOn your PC
⚠️ "runs but slow" means part of the model spills to system RAM/CPU - it works, just slower. ❌ models can still run very slowly on CPU only, but it's not recommended.

3 · Set it up - step by step (desktop)

AGenO runs local models through Ollama (free). It takes about 5 minutes once.

  1. Install Ollama. Go to ollama.com/download, run the installer for your OS, and finish. (It runs quietly in the background.)
  2. Open a terminal. Windows: press Win + R, type cmd, Enter. Mac: Cmd + Space, type Terminal, Enter.
  3. Download your tier's model. Paste the command below (it's already set to the best model your PC can run from the table above) and press Enter. First download is a few GB; after that it's instant. 
    ollama pull hermes3
  4. Let AGenO's website reach it - run this once, then fully quit & reopen Ollama: 
    Windows:   setx OLLAMA_ORIGINS "https://agenoai.com"
Mac:       launchctl setenv OLLAMA_ORIGINS "https://agenoai.com"
Linux:     export OLLAMA_ORIGINS="https://agenoai.com"   (add to ~/.bashrc)
  5. Turn it on. Open AGenO chat, click the 💻 button next to the model menu (it turns green), and send a message. It now runs on localhost:11434 - your own machine.
🔒 In local mode your prompts never leave your computer - AGenO can't even see them - and you skip the queue completely. Each local message still counts toward your plan's weekly local allowance (shown live in chat), so a higher plan still gets you more - you just never wait in line.
⚠️ If chat says it can't reach your local AI: make sure Ollama is running, you ran the OLLAMA_ORIGINS step and restarted Ollama, and you pulled the model. Local mode also needs you to be signed in (so it can meter against your plan).

4 · Want a bigger model than your tier?

Running locally, you're not limited to your tier's size - if your GPU is strong you can bring a higher-parameter model (e.g. a 32B or 72B) for even better answers. That's a separate “bring-your-own-model” option so the website knows it's an intentional upgrade, not a random mismatch.

See the bring-your-own-model add-on →
I keep the model matched to your plan by default so results stay consistent - the BYO option lets you opt into something larger on purpose.

5 · Run image / video / music / voice locally - exact commands

Images, video and music run through ComfyUI (free, open-source); voice uses a TTS engine. Good news - local image, video and music generation are now wired into the AGenO site (beta): set up ComfyUI once, flip the 💻 toggle on the page, and press Generate. Image uses a built-in workflow; for video & music you paste your own ComfyUI workflow once (so it matches your exact models). Copy-paste each block below into a terminal.

Step A - Install ComfyUI once · powers image · photo · video · music
Easiest: one-click app for Windows/Mac from comfy.org/download. Or from a terminal (any OS, NVIDIA GPU): 
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
python -m venv venv

# activate it - Windows:
venv\Scripts\activate
# …or  Mac/Linux:
source venv/bin/activate

# PyTorch (NVIDIA CUDA 12.4 build) + ComfyUI deps:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt
Step B - Start ComfyUI so the AGenO site can reach it (CORS) important
This flag is what lets the website talk to your machine - without it the browser is blocked: 
python main.py --enable-cors-header "https://agenoai.com"
Leave that window running. ComfyUI is now live on localhost:8188. (One-click app? Add --enable-cors-header "https://agenoai.com" to its launch arguments.)
🖼️ Image generation ✅ wired to the site · beta · ~6 GB VRAM
Download the exact model AGenO's local image mode uses (SDXL Turbo, ~6 GB): 
pip install -U "huggingface_hub[cli]"
huggingface-cli download stabilityai/sdxl-turbo sd_xl_turbo_1.0_fp16.safetensors --local-dir ComfyUI/models/checkpoints
That's it. Open the image generator, click the 💻 button (turns green), type a prompt, hit Generate → it renders on your GPU, no queue. (Still counts toward your weekly image limit.)
🎬 Video generation ComfyUI · ~12 GB+ VRAM
✅ Now one-click on the site (beta). Download the exact LTX model the built-in workflow uses: 
huggingface-cli download Lightricks/LTX-Video ltx-video-2b-v0.9.5.safetensors --local-dir ComfyUI/models/checkpoints
Then open the video page, flip 💻, type a prompt and press Generate - no workflow needed. (If your model file has a different name, click “Load default” in the panel and edit ckpt_name, or paste your own Save-(API-Format) workflow.)
🎵 Music generation ComfyUI · ACE-Step · ~8 GB
Add the ACE-Step nodes; ComfyUI fetches the model on first run: 
cd ComfyUI/custom_nodes
git clone https://github.com/ace-step/ComfyUI_ACE-Step
cd ..
# restart ComfyUI, then load the ACE-Step workflow
Easier alternative: ComfyUI Manager → search “ACE-Step” → Install. Then Save (API Format) your ACE-Step workflow and paste it into the 💻 panel on the music page (use %PROMPT% for your tags) to generate locally through the site.
🎚️ Fast and High Quality both run on ACE-Step locally - the 💻 music page picks the right step count automatically. ✨ Full Song (complete ~1.5-min songs with vocals) uses the DiffRhythm model - the easiest way to add it is the one-click AGenO Local dashboard (the ⬇️ Download button at the top of this page) - press “✨ Add Full Song” and it installs the node + weights (~14 GB) into your ComfyUI.
✨ Photo enhancing / upscaling ComfyUI · ~6 GB
Same ComfyUI - add an upscaler model and use an img2img/inpaint workflow: 
huggingface-cli download Kim2091/UltraSharp 4x-UltraSharp.pth --local-dir ComfyUI/models/upscale_models
✅ Now one-click on the site (beta): with ComfyUI running + the SDXL-Turbo model, open the photo editor, upload a photo, flip 💻, describe your edit and press Edit - it runs img2img on your machine. (For big upscales add the model above and use a workflow with an Upscale Image (using Model) node.)
🎙️ Voice - fully local ✅ wired · hear + think + speak on your PC · ~5 GB
Voice runs end-to-end on your machine when you run a small speech server (Chatterbox TTS + Whisper STT) that exposes /tts and /stt on localhost:8123 with CORS enabled. 
pip install chatterbox-tts faster-whisper
Start the server (it must allow cross-origin requests - add Access-Control-Allow-Origin: *), then on the voice page click 💻. Now your speech-to-text (Whisper), the reply (your Ollama model) and the spoken voice (Chatterbox) all run locally - no queue, fully private. Each piece falls back to the cloud automatically if its local service isn't up.
💬 Chat · 🎲 Quest · 📘 Learn helper Ollama · done in Steps 1-5
All text models - already covered by the ollama pull steps above. The Quest game and Learn helper will reuse the very same model.
💡 One ComfyUI install (Steps A-B) covers image, photo-edit, video and music. Keep its terminal running whenever you want local generation. Models are multi-GB downloads - grab them on a good connection.

← Back to the app · See plans & pricing