For the complete documentation index, see llms.txt. This page is also available as Markdown.

🔅Liteforms™

Liteforms™ is a holographic conversational AI character platform.

Liteforms is an open-source browser-based avatar chat app for creating and talking with an interactive 3D character. It renders a VRM avatar in real time, supports typed and spoken conversation, and can send the avatar view to a Looking Glass display through Looking Glass Bridge and WebXR.

Try Liteforms on your browser now

Looking Glass Go support requires Desktop Mode and Looking Glass Bridge.


Features

  • Realtime 3D avatar scene with animation, expression, mouth movement, and lip sync

  • Chat with the character using text or microphone input

  • Character editor for name, pronouns, personality, greeting, and VRM avatar upload

  • Built-in browser-local model options for LLM, text-to-speech, and speech-to-text

  • Support for external LLM providers, including OpenAI, Anthropic, Google, OpenRouter, and OpenClaw

  • Support for external speech providers, including OpenAI, Google, ElevenLabs, and Deepgram

  • Looking Glass WebXR support for viewing the avatar on a connected Looking Glass display

  • Browser-local storage for character settings, provider configuration, credentials, and uploaded VRM files


Requirements

  • A modern Chromium-based browser is recommended, especially for local models, microphone input, and audio playback

  • Microphone permission if you want to use speech input

  • Looking Glass Bridge if using a Looking Glass Go

  • Looking Glass Go must be connected to a computer in Desktop Mode

  • Optional: Node.js 20 or newer and npm if running Liteforms locally from source

  • Optional: OpenClaw CLI and a configured OpenClaw Gateway if using OpenClaw Gateway as the LLM provider

  • Optional: API keys or credentials if using external LLM, text-to-speech, or speech-to-text providers


Setup

If you want to run Liteforms from the web, simply go to the web page and set up your service providers.

If you are running Liteforms locally from source, download or clone the source code from GitHub, install dependencies, and start the development server:

Then open the local URL printed by Next.js, usually:

If port 3000 is already in use, Next.js may choose another port.

Connecting a Looking Glass Go

To use Liteforms with a Looking Glass Go:

  1. Connect the Go to your computer with the USB-C cable.

  2. Make sure the Go is in Desktop Mode, not Standalone Mode.

  3. Install and run Looking Glass Bridge.

  4. Open Liteforms in your browser.

  5. If Liteforms shows a Bridge banner, confirm that Bridge is running and the Go is connected.

Once Bridge is connected, a "Hologram-iphy" button will appear under the character. Press it to enter hologram mode.


First Launch

On first launch, Liteforms asks you to configure the character's model and speech pipeline.

Select LLM

The LLM is the character's "brain." Choose a model provider from the Select LLM screen.

You can use a browser-local model, or connect an external provider. If the provider needs a credential, enter it in the credential field. Credentials are stored locally in your browser and are never shared with us.

Use OpenClaw Gateway

OpenClaw Gateway lets Liteforms use your local OpenClaw agent as the character's LLM. Because OpenClaw Gateway runs locally on your computer, the most reliable path is to run Liteforms locally from source using the instructions above.

OpenClaw's OpenAI-compatible chat-completions endpoint is disabled by default. Liteforms needs that endpoint enabled before it can use OpenClaw Gateway.

Option 1: Ask OpenClaw to set up Liteforms

The Liteforms repository includes an OpenClaw skill at:

OpenClaw skills are normally folders that contain a SKILL.md file. To make the Liteforms skill available to OpenClaw, place the skill folder in one of OpenClaw's skill locations:

  • Global skill location: ~/.openclaw/skills/liteforms-web/SKILL.md

  • Workspace skill location: <workspace>/skills/liteforms-web/SKILL.md

On Windows, the global skill location is usually:

After installing the skill, confirm OpenClaw can see it:

If OpenClaw does not show the skill after you add or update it, start a new OpenClaw session or restart the Gateway:

Then ask OpenClaw:

OpenClaw should verify git, node, npm, and openclaw; clone or update Liteforms; install dependencies; enable Gateway chat completions; start or restart the Gateway; start Liteforms; and give you the local Liteforms URL. If Gateway auth is enabled, OpenClaw may retrieve the Gateway token and place it into Liteforms. Treat that token as a local secret.

Option 2: Configure OpenClaw manually

In an OpenClaw terminal, enable the Gateway chat-completions endpoint:

Then start or restart OpenClaw Gateway:

If your Gateway uses token auth, get the local Gateway token:

Treat this value as a secret. Do not paste it into public logs, tickets, or shared screenshots.

In Liteforms, choose:

  • Model provider: OpenClaw Gateway

  • Model: openclaw/default

  • Base URL: http://127.0.0.1:18789/v1

  • OpenClaw Gateway token: paste gateway.auth.token or OPENCLAW_GATEWAY_TOKEN if Gateway auth is enabled

If you want Liteforms to use a faster OpenClaw response mode, enable fast mode in the OpenClaw session or agent that Liteforms will call.

Select Text-to-Speech

Text-to-speech is the character's voice. Choose a local voice model such as Kokoro, or connect an external voice provider such as OpenAI, Google, ElevenLabs, or Deepgram.

Select Speech-to-Text

Speech-to-text controls microphone transcription. Choose the built-in Distil-Whisper option (English only), or connect an external transcription provider.

After configuration, select Start Liteforms. If local models are selected, Liteforms will show download progress before continuing.


Using the App

Liteforms has two main areas:

  • The avatar scene, where the 3D character is displayed

  • The chat panel, where you edit the character, configure providers, and talk with the avatar

The default avatar is loaded automatically. If a custom VRM avatar was previously uploaded, Liteforms restores it from browser storage.


UI Controls

Character

The Character section is open by default. Use it to edit:

  • Name: the character's display name

  • Pronouns: the character's pronoun set

  • Personality: the system prompt that guides how the character speaks and behaves

If OpenClaw is selected as the provider, character identity is managed by OpenClaw's soul system. Switch to another provider to edit the character persona directly in Liteforms.

Advanced Character Settings

Open Advanced inside the Character section to load a custom VRM avatar.

  1. Select Load VRM.

  2. Choose a .vrm file from your computer.

  3. Wait for the avatar to load in the scene.

Use Reset to return to the default avatar.

The Explore VRoid Hub and Explore OSA buttons open external avatar galleries where you can find VRM-compatible characters.

Settings

Open the Settings section to review the current model and speech configuration.

The Settings section shows:

  • Model provider

  • Model

  • Voice provider

  • Speech input provider

Select Configure to reopen the setup flow and change LLM, text-to-speech, or speech-to-text providers.

Advanced Settings

Open Advanced inside the Settings section to view local model and provider tools.

This area includes:

  • Local model download and cache status

  • Cache usage

  • Clear button for clearing local model cache

  • Test voice button for testing text-to-speech

  • Test STT button for testing speech-to-text


Chat Controls

Use the message box at the bottom of the chat panel to type a message, then select Send.

Liteforms streams the response into the chat and plays speech when a voice provider is configured. The avatar's mouth movement follows the generated speech.

Microphone Input

Liteforms supports three microphone modes:

  • Auto (default): starts listening and sends after a pause in speech

  • Tap: tap once to start recording, then tap again to stop

  • Hold: hold the microphone button while speaking, then release to send

Use the small mic mode control next to the microphone button to switch between modes.

If using a realtime voice provider such as Google Live or OpenAI Realtime, the microphone button starts or stops the live voice session instead of recording a short clip.


Viewing on a Looking Glass Display

When a Looking Glass display is connected through Bridge, Liteforms can send the avatar scene to the display.

  1. Confirm Bridge is running.

  2. Confirm the display is connected and detected.

  3. Select Hologram-iphy to start the Looking Glass WebXR view.

  4. Select Make it boring to exit the Looking Glass view.


Credentials and Privacy

Provider credentials entered in Liteforms are stored in browser-local storage. They are not committed to the repository and are not saved to a hosted account system by default.

Local model files are stored in the browser cache. You can clear them from the Advanced Settings section.

Uploaded VRM avatars are stored locally in the browser using IndexedDB when available.


Troubleshooting

Bridge banner appears

Make sure Looking Glass Bridge is installed, open, and running. If using a Go, confirm that it is connected to the computer and in Desktop Mode.

The Go does not show the Liteforms view

Confirm that the Go is in Desktop Mode. Standalone Mode is used for the mobile app, Blocks sync, and playlists, but not for Liteforms.

Local models are slow to start

The first launch may download local model files. Wait for the model progress step to finish. On slower connections this can take several minutes.

Microphone does not work

Check that the browser has permission to use your microphone. A Chromium-based browser is recommended for the most reliable microphone and audio behavior.

Speech or chat provider fails

Open Settings, select Configure, and confirm the selected provider, model, endpoint, and credential. If using a provider API key, make sure the key is active and has access to the selected model.

OpenClaw Gateway returns 404 or 401

If Liteforms reports a 404 for /v1/chat/completions, rerun the OpenClaw chat-completions config command and restart the Gateway.

If Liteforms reports a 401, confirm Gateway auth is enabled and re-enter the correct Gateway token from gateway.auth.token or OPENCLAW_GATEWAY_TOKEN.


Support

For Looking Glass hardware or Bridge issues, contact Looking Glass support at [email protected].

For developer questions related to Looking Glass WebXR or Bridge integrations, contact [email protected].

Last updated