Ami: Your 3D Chatbot
A downloadable game for Windows
What is Ami (ah-mee)?
Ami is your Unreal Engine Agent, powered by private/open-source, as well as commercial/closed-source- Large Language Models. Ami will roam around the house doing various things while you chat and interact. Ami is continuously evolving and grows further into full ai agent capabilities with each new update .
- Ami uses your OpenAI and ElevenLabs API key to facilitate more humanized LLM interactions, all prepackaged for the more casual AI user.
- Ami supports whatever model you can find and run locally using LM Studio or Ollama. These local models allow for free, private and uncensored chats. Local models can be hosted on any PC on your local network to use another GPU for LLM inference while you run Ami on another PC.
- Ami has a chat history and remembers your interactions as long as you want.
- Chat history is exportable as an external JSON file.
- Ami supports Text-To-Speech for OpenAI, ElevenLabs and Offline Text-To-Speech (private).
- Optional female or male avatars and 18+ only animations.
- OpenAI API and Patreon members have access to asking Ami to web search during chats.
- OpenAI API users have access to file uploading and searching, allowing them to chat about the contents of documents given.
- 3 different environments for your Ami: an empty basic plane, a simple Starter Home and a furnished Triplex.
- VR SUPPORT for interacting with a companion via OpenAI (GPT), LM Studio, Ollama, or the Robot Punch API (available to Patreon members).
- Push to Talk support for OpenAI API users and Patreon members.
- When using the OpenAI API or you are a Patreon Member, your Ami can sometimes initiate chats with you first, by being able to speak up first without being first prompted by the user.
- Patreon Members get access to build in Text-To-Speech using the Elevenlabs API.
Why use an interface like Unreal Engine for a chat-bot?
- To humanize the interaction and create a more emotional connection.
- For a more powerful and immersive graphical engine, permitting 3D companions, rendered in high graphical fidelity or Virtual Reality.
- To create a digital companion that does not require a subscription service. I pledge to support open-source models and free-to-use chatbots.
- To utilize Unreal Engine's game save feature and various plugins to store private chat history locally, for completely offline and persistent LLM interactions.
- To improve virtual companion accessibility to users of all experience levels with fully uncensored and private interactions.
Ami is a work in progress.
Ami (like AI) is evolving at a rapid pace. Thanks for your patience and support as I work to improve a virtual companion experience for everyone.
Upcoming Features in Development
- Ami orients head position to look at user/camera.
- Lip sync for your companions text-to-speech.
- Ami can see the user's 3d avatar from their own perspective in-game, utilizing vision models and in-game cameras.
- Talk to your Ami while you're away, via Discord.
Download
Click download now to get access to the following files:
Development log
Comments
Log in with itch.io to leave a comment.
Hi I'm wondering if I could get any help. I've tried using the PC version but after making a character the camera is just stuck on the bottom of the map, and in the VR version I can't seem to click on anything in the main menu.
The camera appearing in the floor sounds like when the game boots to the HMD but you’re viewing on desktop. It might be launching in VR if the Quest HMD is in link mode.
In VR, currently, you are intended to hit the pause button and open your wrist menu, which enables your VR pointer to click in menus and stuff.
On the Quest, hitting B should open the interaction menu once you’re inside- and if you’re using GPT or the Robot Punch api, holding Y will handle short PTT recordings.
I might have got those buttons mixed up.
I should make a video explaining some things, I think.
sorry but my dumb brain cant make sense of the key stuff can you give me a rundown? are the keys required?
No worries, the API key is the toughest part here, so it makes sense.
I'll make a detailed setup video soon, but the gist is you request an API key from OpenAI or run LM Studio somewhere on your main PC or home network.
If you request an API key, you'll also have to add whatever amount in funds to the account- so this is the money that is consumed when you make API calls and it's something like a penny per message, or now that I'm adding Vision, the requests for image analysis are maybe $0.03 each. But talking to GPT like this is far cheaper than a $15 month subscription and I'm trying to get it it match the functionality you can get through your usual GPT interaction. The API key needs to be saved into a specific spot on your PC that my software looks at to read from. I will add a mechanism to setup this key at the required location for you, but I need to be sure not to break an existing key that is properly setup with it all the same. So I'll come along to help with the API issue soon for more users.
If you're using LM Studio or want private offline conversations, you need a local PC powerful enough to run LM Studio + whatever model you choose from the software (you'll find various models that specialize in different tasks, like coding or roleplaying to choose from). With LM Studio running a local server, Ami needs an IP address to connect to and you can inference with the local LLM through Ami.
The Elevenlab API key is optional, but for people who have Elevenlabs, you add the API key the same way you do the OpenAI key. As long as the system variable holding the API key is named what OpenAI and Elevenlabs tell you to name it, my software will find the key all the same.
But if you don't have OpenAI keys or Elevenlab keys, you can still use LM Studio + Offline TTS.
After Ami makes enough money, I can setup a server and establish a connection on behalf of the user in a fashion that requires no API key. But until I can afford the servers, this seems the best I can do to make it as cheap and accessable as possible.
So it'll get easier eventually, but for now it's a little technical and I'll try to help with that fact.