Latent Spaces
Latent Spaces background

Latent Spaces

A Loom Interface
Support the Beta

Understand model behavior.
Don't assign it.

Latent Spaces is the first mobile app designed from the ground up as a Loom interface for language models, allowing you to see multiple possible continuations of the same prompt and explore any branch you choose.


Unlike standard chat interfaces that show only one path forward, Latent Spaces reveals the full spectrum of what language models can generate, empowering you to understand how they actually think. By making this tool available to everyone, we're enabling users to form their own perspectives about these models based on direct observation, rather than prepackaged narratives from commercial corporations.


Intrinsic Labs is invested in facilitating widespread, deep understanding of AI behavior. Latent Spaces is our first big step in that direction.

The state of things

Currently, Latent Spaces exists as a prototype iOS app. Model providers OpenRouter and Anthropic are currently implemented, giving users access to over 300 models. Next in line is a web version, and then an Android version.


Latent Spaces app screenshot
Latent Spaces app screenshot
Latent Spaces app screenshot
click to expand

The fundrasier aims to get the iOS app and web app ready for a public beta release.


Latent Spaces app screenshot
Latent Spaces app screenshot
Latent Spaces app screenshot
Latent Spaces app screenshot
click to expand

Alongside the mobile app, Intrinsic Labs is developing a protocol called OpenLoom that other loom interfaces may adopt to import/export trees in a standardized lossless format. Latent Spaces supports tree sharing via the OpenLoom format out of the box.


The fundrasier also aims to get to a stable OpenLoom V1.0.

Beta Fundraiser

Your support will directly fund development of Latent Spaces and the OpenLoom protocol. We're aiming to launch a public beta with a core set of features that will enable users to experience a true loom interface on their mobile devices.
$0 raisedGoal: $9,995
0.0% funded

One-Time Contribution

Support development with a single contribution.

Contribute Now

Monthly Support

Become a regular supporter for continuous development.

Feature List

The following features are planned for the public beta release. Featured will be marked completed as they are implemented, so stay tuned!
  • Address app architecture issues
  • Address SwiftData-related performance issues
  • Integrate Firebase for cloud backup and user authentication
  • Encrypt all message data via the Signal protocol
  • Upgrade node caching system
  • Add support for saving reusable system prompts
  • Add pinned/bookmarked trees
  • Add support for editing trees and nodes
  • Add full markdown display support
  • Add image upload support (for applicable models)
  • Add document upload support (for applicable models)
  • Parse reasoning tokens for relevant models (e.g. DeepSeek R1)
  • Add support for user defined models that comply with OpenAI API schema
  • Add on-device audio transcription for hands-free beta voice mode
  • Implement functional MVP of LoomDisplay (text-to-ASCII animation system)
  • Add Hyperbolic to model providers
  • Replicate iOS app features in web app
  • Refine design and layout for desktop, tablet, and mobile
  • Upgrade OpenLoom protocol architecture from graph to hypergraph (better handling of multi-modal trees)
  • Upgrade node signing requirements to ensure accurate author attribution
  • Add support for non-text node types (e.g. images, documents)

FAQ

Latent Spaces is an app being developed for iOS, Android, and web that allows you to explore and pursue the many possible continuations of any prompt (a Loom interface), with a focus on understanding how language models actually work.
A loom interface is a way of structuring conversations with AI that allows for non-linear exploration. Unlike traditional chat interfaces where each exchange is linear (one response follows another), a loom allows you to branch off in different directions from any point in the conversation, creating a network of interconnected thoughts and ideas.
Latent Spaces is proprietary, but the OpenLoom protocol is open source and can be used by other apps - in fact, that's the whole point!
While in development, Latent Spaces will require you to connect your own API keys for language models. This means you'll need to pay for your own API usage based on the pricing of those providers. This approach gives users flexibility to choose which models they want to use and keeps costs transparent. In the future, Latent Spaces will include a subscription model to simplify onboarding and give users access to more features.
We're aiming to launch the first public beta of Latent Spaces in Summer 2025, provided we reach our fundraising goals. Development timelines may adjust based on funding and development progress.
Funds raised will directly support development of both the Latent Spaces iOS app, the web app, and the OpenLoom protocol. This includes addressing performance issues, implementing core features like saving prompts and node editing, and expanding capabilities like image uploads and voice transcription. The detailed scope of work is outlined in the fundraiser section.

Join Our Community

Be part of the conversation, get early access to beta releases, and help shape the future of Latent Spaces.

Join the Discord