AppJuice User Guide

Apps are the center of AppJuice.

AppJuice treats an app as the durable result: an interface, local data, runtime behavior, and optional AI that remain available after the building session ends.

Build and run in one platform: Describe the workflow, build the app, and run it immediately. AppJuice supplies the runtime instead of leaving the result as a project that still needs to be deployed elsewhere.
Connect AI at the platform level: Apps can use configured AI providers, built-in AI capabilities, or a Runtime Agent without managing provider credentials in each app.
Combine capabilities by task: One app can use different providers and models for text, images, speech, media, search, extraction, or multi-step agent work.
Augment the user's workflow: The app keeps its data, controls, and decisions visible while AI assists with the parts that benefit from generation, analysis, or automation.
Use, Tweak, and keep: Run the app for normal work, improve it in context, and continue with the same app and saved state.

AppJuice desktop app in Cards view — **Cards view.** See app descriptions and open Use or Tweak directly from each card.

AppJuice desktop app in Compact view — **Compact view.** Scan and open apps quickly when the library grows.

AppJuice desktop app in List view — **List view.** Review app status, descriptions, and actions in a denser layout.

Platform features

AppJuice supplies the runtime and controls that apps share. Each app keeps its own interface and workflow.

Local app runtime: Apps, sessions, private data, and generated files are stored on your computer by default.
Use and Tweak: The working app and its editing workflow stay together.
AI service routing: Configure cloud providers, gateways, and supported local model servers once for use by apps.
Built-in AI capabilities: Apps can request text, image, speech, music, video, search, and extraction features through shared platform APIs.
Building skills: Builder uses AppJuice's built-in instructions, optional global skills, and skills installed only for one app.
Agent Runtime: Apps can run multi-step agent sessions with approved tools, files, and generated artifacts.
Controlled system access: Folder approvals, sandboxed commands, native bridges, and app permissions limit what an app can reach.
Platform Vault: App backends can use saved service credentials without placing secret values in the browser interface.
Remote Access: Selected apps can be opened from an approved browser on your network or through Tailscale.

Set up your first app

Start with one task that has a clear input and result.

Connect an AI service
Open Settings and connect a cloud provider, model gateway, or supported local model server.
Create or install an app
Describe a workflow to Brew, import an app, or install one from Juicery.
Use the app
Add the files or information it needs, review the result, and save or export your work.
Make a change
Switch to Tweak and describe the improvement you want.

Note: The standard workflow does not require a terminal, repository, deployment step, or code review.

Create an app with Brew

Start by describing the problem in your own words. Brew takes the role of a product designer: it asks focused follow-up questions, often with a few clear options, so you can work through the audience, workflow, outputs, and important boundaries together.

When the idea is clear enough, Brew summarizes what you agreed and prepares an initial product plan for you to review before Builder starts.

Brew product-design conversation with selected and available workflow choices

Use skills while building an app

Skills give Builder extra guidance while it creates or changes an app. This section is about building skills; skills used by an app after it is running are configured separately.

Built-in skills that come with AppJuice

AppJuice includes guidance for the things its apps commonly need, such as AI features, safe access to files, and a clear user experience. You do not need to install or manage these skills. Builder uses the relevant guidance automatically.

Building settings showing Global Skills and seven built-in AppJuice skills

Skills you add for every app

If you want Builder to follow the same specialized guidance across many apps, install the skill once in Settings → Building. It will then be available whenever you create a new app or Tweak an existing one.

Search Global Skills dialog with design skills available to install for every app

Skills you add for one app

Some guidance is useful for only one app. Open that app in Tweak, choose Skills, and install it there. Builder will use it while working on that app without adding it to every other app you build.

Skills for Building This App dialog with one app-only skill and a search field

Use and edit the same app

Use mode is the normal working interface. Tweak mode changes that app without moving the work to a separate development tool.

Work in Use mode

Use your own records, files, controls, and saved app state. Note any missing filter, export, approval step, or unclear result while doing the task.

Describe the change in Tweak mode

State what should change and any behavior that must stay the same. After reviewing the update, return to Use mode.

Nectra reading app with Tweak mode open beside the running app

Configure AI services

Provider connections, platform capabilities, and model preference are configured separately.

Connect providers and local models

AI Services can contain several cloud providers, compatible gateways, and supported Ollama or LM Studio servers. Credentials remain in platform-managed storage.

AI Providers settings with connected cloud providers and a local Ollama server

Use built-in AI capability routes

AppJuice provides a shared set of routes for widely used AI capabilities, including text and image understanding, image generation, speech, media, search, and extraction. The platform connects each route to suitable models from the AI providers you have configured.

An app can call these capabilities through stable AppJuice backend APIs instead of integrating every provider separately. The same capabilities can also be made available as Tools for a Runtime Agent or Builder, so an agent can use them as part of a longer task. AppJuice handles provider credentials, model availability, and the preferred model order in one place.

The current version supports the capabilities below. A route is ready when a suitable provider or backend has been connected. More capabilities may be added as AppJuice expands its provider support.

Builder / Runtime Agent
Text
Image understanding
Image generation
Speech to text
Text to speech
Music generation
Video generation
Web search
Web extraction

AI capability list with Builder, Runtime Agent, text, image, and speech routes

Set the model order for each capability

Each capability has its own ordered list of ready models. Drag models into the order you prefer; AppJuice uses that order when an app or agent does not request a specific model. You can reset the list to the platform default at any time.

Image Generation capability with six ready models in a configurable order

Run agents inside an app

Agent Runtime supports work that takes several steps or needs approved tools, files, session history, or generated artifacts.

Single-agent Runtime

AppJuice provides three ways to add one Runtime Agent to an app. Choose the level that fits how the agent should appear and work in the app.

Full conversation UI — AgentConversation: A ready-to-use conversation area for apps where the agent is a main part of the experience.
Expandable chat bubble — AgentChatWidget: A floating assistant that stays minimized in the corner until the user opens it. It suits apps where the agent supports the main experience without taking up a permanent panel.
Custom Agent APIs: The most flexible option. An app can use its own interface, start agent work from buttons or workflows, or connect the Runtime Agent to its backend.

Nectra reading app with an in-app Runtime Agent discussing the current chapter

App-defined agent orchestration (Testing)

An app can coordinate several Runtime Agents through its own code. The app defines who does each part of the work, when agents run in sequence or in parallel, how results are passed between them, and how the final answer is assembled.

AppJuice also provides an Agent Collaboration View component that apps can use to show the agent organization, working relationships, and live progress without building that interface from scratch.

All Minds app with four specialist agents and a Decision Chair

Other backend APIs

Direct AI helpers: Direct text, image, speech, music, video, search, and extraction helpers for tasks that do not need an agent.
provider-access: Server-side access to configured providers or Platform Vault credentials.

Store service credentials in Platform Vault

Platform Vault stores credentials for services used by an app backend. AI model providers remain in AI Services.

Vault values are encrypted locally, are not shown again in Settings, and are not sent directly to the app's browser interface. An app requests a saved credential through the platform runtime when it needs to call that service.

Platform Vault settings with secure storage, templates, and saved credentials

Files, permissions, and recovery

Apps and their private runtime data are local by default. A cloud AI request still sends the context required for that request to the provider you selected.

Approved paths: You choose the folders an app can read or write.
Controlled system actions: Local commands and native features run through platform controls, app permissions, and sandboxing.
Local inference: Supported Ollama or LM Studio models can keep model inference on the same machine.
Install checks: Imported and shared apps are scanned so capability and security findings can be reviewed.
Private sharing boundary: Normal app sharing excludes provider keys, private runtime data, sessions, and local files.
Version history: You can inspect changes and restore an earlier app version when a Tweak does not work as intended.

Open selected apps from another device

Remote Access makes selected running apps available to approved browsers. Use the local network nearby or a private Tailscale connection while away.

Per-app access: Only apps you select are available remotely.
Browser approval: Approve each browser and choose Use only or Use and Tweak access.
Direct connection: AppJuice does not relay the app through an AppJuice cloud service.

Remote Access settings with device authentication and shared app controls

Install and share apps

Install from Juicery: Review an app, install it locally, and connect it to your own AI setup.
Share or remix: Share the reusable app source and recipe without including your keys, files, or private sessions.

Common questions

Which AI provider can I use?

Open Settings, then AI Services. You can connect supported cloud accounts and APIs, compatible model gateways, or supported local servers. Available models depend on the services you configure.

Can I keep model inference local?

Yes, when the app's workflow and capability are supported by a connected local model such as Ollama or LM Studio. A workflow that uses a cloud service sends the required request context to that service.

Does every app need an agent?

No. A simple action can call a built-in AI capability directly. Runtime Agent is intended for multi-step work, tool use, files, or generated artifacts.

What is the difference between Builder skills and Runtime Agent skills?

Builder skills guide AppJuice while it creates or changes an app. Runtime Agent skills belong to an app's agent and are available when that app performs user tasks. Installing a global or app-only Builder skill does not automatically add it to a Runtime Agent.

Can one app use several Runtime Agents?

Yes. The platform runs each configured agent while the app defines the collaboration pattern and failure handling.

Can I open an app from my phone?

Yes. Enable Remote Access, select the app, and approve the phone's browser. Use the same local network or a configured Tailscale connection.

Will sharing include my keys and files?

Normal app sharing carries the reusable app, not provider credentials, private runtime data, sessions, or local files. The recipient connects the app to their own services.

AppJuice user guide