Connect Your Own AI

🔌 Connect Your Own AI

In the browser and on Android, Formamorph starts on the Demo AI. It is a small free model, and it is there so you can try the app with no setup. The desktop app starts on its Built-In Engine instead.

The AI writes everything you read. A stronger model gives you sharper narration, a better memory of your story, and entities that act the same from turn to turn. Nothing else changes the experience as much.

Every route ends in the same place: Settings → Endpoints → Text. The Endpoints tab shows in Simple and Advanced mode.

How to Connect LM Studio

1. Download LM Studio from lmstudio.ai and install it.
2. Open the Discover tab and download a model.
3. Open the Developer tab.
4. In the server settings, turn on Enable CORS. Your browser blocks the connection without it.
5. Turn on the Start server switch.
6. Load your model with the model loader. The Developer tab shows the server address, such as http://localhost:1234.
7. In Formamorph, open Settings → Endpoints → Text.
8. In the Preset list, select Add New Preset….
9. Type a name in New Preset.
10. Select Save.
11. Paste the server address into Endpoint URL.
12. Leave API Token empty.
13. Type the model identifier that LM Studio shows into Model Name.

How to Connect Ollama

1. Download Ollama from ollama.com/download and install it.
2. Download a model: ollama pull <model>. Use a model name from the Ollama library.
3. Set the environment variable OLLAMA_ORIGINS to *. Your browser blocks the connection without it. See OLLAMA_ORIGINS on Each System.
4. In Formamorph, open Settings → Endpoints → Text.
5. In the Preset list, select Add New Preset….
6. Type a name in New Preset.
7. Select Save.
8. Type http://localhost:11434 into Endpoint URL.
9. Leave API Token empty.
10. Type the model's name into Model Name, exactly as ollama ls lists it.

How to Connect a Hosted API

1. Make an account with a service that offers an OpenAI-compatible chat-completions endpoint.
2. Get an API token from the service. Some services call it an API key.
3. Open Settings → Endpoints → Text.
4. In the Preset list, select Add New Preset….
5. Type a name in New Preset.
6. Select Save.
7. Paste the service's chat-completions URL into Endpoint URL. It usually ends in /v1/chat/completions.
8. Paste your token into API Token.
9. Type the identifier of the model you want into Model Name.

How to Use the Desktop Engine

1. Get the desktop app from formamorph.ai and install it.
2. Open Settings → Endpoints → Text.
3. In the Preset list, select Built-In Engine.
4. Select Manage Models…. The Local model dialog opens.
5. Open Recommended. It groups models by the GPU memory they need, and starts on the group for your GPU.
6. Select Download on a model.

With Auto-Load on, the default, the model loads when its download finishes. With it off, select Load on the model.

How to Play Against Your PC from Another Device

1. Make your server accept connections from your network. In LM Studio, turn on Serve on Local Network. In Ollama, set OLLAMA_HOST to 0.0.0.0:11434.
2. On the other device, open Settings → Endpoints → Text.
3. Type your PC's network address into Endpoint URL, such as http://192.168.…:1234.

In the browser, Chrome may block a public page from reaching your own network. The Android app has no such block.

---

🧭 Pick a Route

Route	You need	Best when
🖥️ Run a local server	A PC with a capable GPU, plus LM Studio or Ollama	You want to play in the browser or on mobile, with the model on your own PC
☁️ Use a hosted API service	An account with a service, and its API token	Your hardware can't run a good model
📦 Use the desktop app	A PC with a capable GPU	You want the fewest installs

🎯 Which Model to Pick

• Pick a model tuned for roleplay or conversation.
• As a rough guide, use 12B parameters or larger.
• Past that, use the largest model that your hardware runs well. A model that does not fit in your GPU's memory runs slowly.

The Text Endpoint Fields

Field	What goes in it
Preset	The saved endpoint you use. Add New Preset… makes a new one.
Endpoint URL	The server address. A bare address such as http://localhost:1234, or one that ends in /v1, gets /v1/chat/completions added. The line under the field shows the full URL. Any other path stays as you typed it.
API Token	The service's token. Leave it empty for a local server.
Model Name	The model identifier, exactly as the server shows it.

💡 If the connection fails, select Trouble Connecting? under Endpoint URL. It opens a checklist of the usual causes.

OLLAMA_ORIGINS on Each System

How you set the variable depends on your system:

System	How
Windows	Quit Ollama, edit the environment variables for your account, add OLLAMA_ORIGINS, then start Ollama again
macOS	Run launchctl setenv OLLAMA_ORIGINS "*", then restart the Ollama app
Linux	Run systemctl edit ollama.service, add Environment="OLLAMA_ORIGINS=*" under [Service], then run systemctl daemon-reload and systemctl restart ollama

The Desktop Engine

The desktop app has the AI engine built in, so there is nothing extra to install.

⚠️ The model still runs on your hardware. The desktop app removes the separate server install. It does not change which models your PC can run. That depends on your GPU and its memory.

The Local model dialog has three views: Installed, Recommended and Options. Recommended groups models as ≤4 GB, ≤8 GB, ≤16 GB and No Limit.

The desktop app also connects to a local server or a hosted service. The steps are the same as above. The browser and Android builds have no Built-In Engine.

The Set Up Your AI Dialog

On the desktop app, a Set up your AI dialog opens when the engine has no model to run. It also opens when you enter a world and the AI can't be reached.

Part	What it does
Detected line	Shows your GPU, its memory and its class
Recommended model	One model that fits your GPU, with a download button
Show all models	Lists every model under Best for Your GPU, Also Fits and Too Big for Your GPU
Open Settings	Opens Settings → Endpoints, to use a server or a service instead
Later	Closes the dialog. You can set this up any time in Settings → Endpoints.

When the model is ready, the title reads You're ready, and Start Playing closes it. For a custom endpoint that fails, Try again checks it again.