Your app, not a black box.
Understand the plan and inspect the code changes. Model Roll fits into your repository and review process.
Local AI, without the infrastructure work
Local AI made easy for builders.
Add local AI to the product you’re building. We’re making the integration, device compatibility, and updates easier, so you can focus on what makes it yours.
In development. Built for developers and AI-assisted builders.
YOUR NEXT FEATUREPlanned experience
Add search by meaning
to my notes app.
A clear plan. Reviewable changes.
- Your product
- A web app
- The capability
- Search & Match by Meaning
- Where AI runs
- Your user’s device
- What you review
- Integration + tests
Your repository. Your pull request.
Your decision to ship.
Your decision to ship.
One product.
Across your platforms.
- Web
- iOS
- Android
- macOS
- Windows
WHAT YOU CAN BUILD
Start with a feature.
Not a model file.
Choose what your product needs to do. Model Roll is being built to handle the local AI underneath it.
Planned launch capabilities. Exact device, model, and language support will be published after testing.
01
Speech to Text
Turn spoken words into something useful.
Add transcription, captions, or voice input to your application. Process supported audio on the user’s device.
Developer detail: speech recognition and transcription.
EXAMPLE USE CASE
A meeting app that turns a recording into an editable transcript without uploading the audio for transcription.
02
Search & Match by Meaning
Find the right result, not just the right word.
Help users find related notes, documents, or products even when their search uses different words. Your app decides which content is searchable.
Developer detail: embeddings, semantic search, and similarity matching.
SEARCH IN YOUR APP
“refund for a damaged order”
A MATCH BY MEANING“The customer asked for their money back because the package arrived broken.”
Illustrative content, not a live search.
03
Understand Images
Make images part of the workflow.
Identify supported objects, remove a background, or select a subject or region inside your app. The user doesn’t need to open a separate editor.
Developer detail: image classification, detection, and segmentation.
EXAMPLE USE CASE
A product-photo tool that removes the background before placing an item into a listing template.
04
Read Documents & Extract Data
Give your app the details it needs.
Read text from supported documents and pull out useful fields. Turn a supported receipt or invoice into information your product can work with.
Developer detail: OCR and structured data extraction.
EXAMPLE USE CASE
An expense app that drafts the merchant, date, and total from a receipt for the user to check.
05
Understand & Organize Text
Less sorting. More useful information.
Summarize a note, label feedback, or extract specific details. Add a defined text task without building a general-purpose chatbot.
Developer detail: summarization, classification, and structured output.
EXAMPLE USE CASE
A research notebook that summarizes a saved article and suggests a category, ready for the user to review.
06
Translate Text Offline
Help users across languages, on device.
Translate text between supported language pairs after the required model download. Useful when connectivity, privacy, or offline access matters.
Developer detail: on-device text translation. Language pairs are validated individually.
EXAMPLE USE CASE
A travel app that lets users translate saved text while they’re offline.
07
Find & Hide Sensitive Details
Build a review step before sharing.
A ready-made workflow that combines supported capabilities to flag details such as email addresses or phone numbers for removal. The user reviews the result before sharing.
Developer detail: sensitive-data detection and redaction. Not a guarantee of complete anonymization.
EXAMPLE USE CASE
A document-sharing app that highlights detected contact details and lets the user confirm what to hide.
FROM YOUR REPO TO YOUR PRODUCT
Five steps.
You stay in control.
The planned workflow turns a feature request into changes you can understand, test, and review.
-
01
Connect your repo
Choose the repository you want Model Roll to work with.
-
02
Pick a capability
Tell us what you want your app to do and where it needs to run.
-
03
Check the fit
See the compatibility plan, requirements, and any unsupported paths.
-
04
Prepare the changes
Get the integration and tests together in a separate branch.
-
05
Review and ship
Review the pull request. You decide when to merge and release.
BUILT FOR THE WAY YOU BUILD
Hands on.
Or agent assisted.
You shouldn’t need to become an on-device AI specialist to add one good feature.
Building with Codex or Claude Code? The plan includes clear documentation and tools your coding agent can use. Prefer to work directly? You’ll have the same control through the dashboard, SDKs, API, and command line.
Integration tools are in development, not yet available to install.AI on your users’ devices.
Local processing is the default. Model files download before first use; supported tasks then run on device.
The work doesn’t stop at install.
Compatibility fixes, model updates, technical health checks, and a path back to a known-good release are part of the product being built.
A FEW THINGS WORTH KNOWING
Clear from
the start.
Can I use Model Roll today?
Not yet. Model Roll is in development. This page describes the planned product, not a released SDK or working integration service. If you’re building something that needs local AI, tell us about it.
What does “local AI” mean?
The AI task runs on your user’s own device instead of sending each request to a cloud AI provider. The model normally needs an initial download. Account access, model downloads, and managed updates may still need an internet connection.
Which platforms will you support?
We’re targeting web, iOS, Android, macOS, and Windows for the public release. Each capability will have its own tested requirements. Exact browsers, devices, models, and language support will be documented before launch.
Is this only for experienced developers?
No. Model Roll is being built for developers, indie builders, small teams, and people building with AI coding tools. The simple path explains what happens in plain language, while the technical details stay available when you need them.
Can I bring my own model?
That’s part of the plan for compatible models and supported custom adapters, including LoRAs. They still need to fit a supported task, runtime, licence, and device. Training and fine-tuning models are not part of the first release.
How will pricing work?
Model Roll is planned as a paid subscription for integration and ongoing operations. Prices and limits will be published before launch. Customer-owned compute, storage, and build systems are the default where they preserve the experience; any separate provider costs will be explained.
Will cancelling break my users’ apps?
No. The planned licensing approach does not remotely disable already-shipped local AI or downloaded models. New managed integrations, releases, and ongoing services depend on an active subscription. Full licence terms will be published before purchase.
BUILDING SOMETHING WITH LOCAL AI?
We’d like to hear
what you have in mind.
Your app, the feature you need, and the platforms you’re targeting. That’s a good place to start.