Skip to content
Projects

acara-app/plate has 40 GitHub stars and 16 forks. GitHot measured +2 stars in 24 hours. Last checked 2026-11-15 15:13 UTC. Last push 2026-11-10.

How these counts are measured

Measured star growth

in 24h
+2
in 7 days
Collecting history
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

Open-source health AI agent platform for self-hosted health questions, personal health data, personalized Nutrition for Metabolic Health, AI meal planning, Telegram chat, and Apple Health sync.

AI

README.md

Original

Acara Plate - Health AI Agent Platform

Acara Plate: Open-Source Health AI Agent Platform

License: O'Saasy Tests Why PHP

Try Acara Plate | Read What Acara Is | Get Acara Health Sync on the App Store

Acara Plate is a self-hostable health AI agent platform for asking health questions, connecting personal health data, tracking daily signals, and turning that context into practical guidance. Nutrition, diabetes support, meal planning, coaching, scanning, the Telegram chat interface, and Apple Health sync are features inside the broader Acara health AI system.

[!IMPORTANT] Acara Plate is informational and educational software, not a medical device. It does not diagnose, treat, or replace a physician, registered dietitian, diabetes educator, or emergency care. Verify AI-generated nutrition guidance before using it for health decisions.

Why We Built Acara

Most people still have to fight their way through ad-heavy search pages, closed portals, or paywalled articles just to get a straight answer about their health. Seeing a clinician can take days or weeks, and many languages, regions, and everyday questions never show up in mainstream tools at all.

Acara exists to make practical, science-aware health guidance easier to reach at the moment someone is actually worried, curious, or trying to change a habit.

Acara Plate is the part you can run yourself. It pulls questions, health data, and day‑to‑day check‑ins into one place so the AI agent can answer with the context of a real person instead of a blank, one-off chat window.

For developers, Plate is a working reference: a Laravel app that shows how to build health agents, ingest device and manual data, design safety-aware prompts, ship a PWA and Telegram chat interface, sync Apple Health, run image analysis, and coordinate structured health workflows around a privacy-first product.

How the Agent Helps

Inside Plate, the AI agent acts as a health assistant that can:

  • Answer health questions in plain language about symptoms, conditions, medications, vaccines, nutrition, mental health, habits, and “what should I do next?” style decisions.
  • Use only the personal context you choose to share, including glucose, sleep, activity, weight, blood pressure, medications, meals, and free-text lifestyle notes.
  • Support healthier routines for sleep, stress, hydration, movement, and daily habits through the AI Health Coach.
  • Help with food decisions through the AI Nutritionist, AI Meal Planner, grocery lists, recipes, and glucose-aware food guidance.
  • Pull in Apple Health data from iPhone via Acara Health Sync, the companion iOS app available on the App Store.
  • Work from Telegram as a chat and logging interface, using the supported bot and the Telegram health logging tool to turn natural language like “fasting glucose 102” or “took 10 units insulin” into structured entries.
  • Scan and interpret meals using public free tools and image-analysis workflows to estimate nutrition and make sense of food choices more quickly.
  • Stay safety-aware by directing clearly urgent messages toward the relevant local emergency number instead of treating the chat as medical care.
  • Work across languages so people are not locked to English to use the system.
  • Keep data under the user’s control, with consent-based health data connections and the option to self-host so developers, families, and clinics can decide where sensitive data lives.

Ecosystem

Project Purpose
Acara Plate The open-source, self-hostable Laravel web app, PWA, health AI agent, health data platform, and API backend.
Acara Health Sync Native iOS companion app that reads Apple Health with permission and sends encrypted health data directly to a user's Plate instance.
Acara Core Premium Acara Cloud and private modules, including long-running semantic memory for preferences, goals, relationships, and context. It is not required for community Plate installs.

Product Links

Use these if you want to understand the public product before reading code:

For Developers

Acara Plate is a Laravel 13 application with a React/Inertia frontend and a comprehensive health AI agent built on Laravel's first-party AI SDK.

Core stack:

  • PHP 8.4 or 8.5
  • Laravel 13
  • Laravel AI SDK
  • Laravel Sanctum, Fortify, Cashier, Socialite, Wayfinder, and Livewire
  • Inertia React 19
  • Tailwind CSS 4
  • PostgreSQL for app data, cache, queues, and sessions
  • Laravel Reverb for WebSocket broadcasting (Redis is used only where required, such as resumable AI streaming)
  • Pest, PHPStan, Pint, Rector, TypeScript, Prettier, and oxlint for quality checks

Architecture

The architecture is shaped by three constraints: agent responses take tens of seconds of model and tool calls, clients expect token-level streaming while they wait, and a privacy-first health product has to stay simple enough for one person to self-host. Plate answers them with a deliberate monolith built around a single hard boundary — synchronous request handling versus asynchronous agent execution. The web tier does the minimum on-request (authenticate, persist, enqueue) and stays stateless; queue workers absorb all AI-provider latency and stream results back over WebSockets; PostgreSQL holds every fact the system depends on, keeping health data transactionally consistent. Every infrastructure role — queue, cache, WebSockets, object storage — is a Laravel driver binding, which is why the same codebase runs unchanged from a laptop to a managed platform.

flowchart LR
    subgraph clients["Clients"]
        web["Web app + PWA<br/>Inertia · React 19"]
        ios["Acara Health Sync<br/>iOS · Apple Health"]
        tguser["Telegram user"]
    end

    subgraph plate["Acara App"]
        subgraph entry["Entry"]
            https["HTTPS entry<br/>TLS termination · load balancing"]
        end
        subgraph application["Application"]
            app["Web tier<br/>Laravel 13 · stateless"]
            q[["Job queue"]]
            workers["Queue workers<br/>AI agent jobs · imports"]
        end
        subgraph state["Data & realtime"]
            db[("PostgreSQL<br/>source of truth · pgvector")]
            cache[("Cache<br/>Redis-compatible · resumable AI streams")]
            ws["Reverb<br/>WebSocket broadcasting"]
        end
        s3[("Object storage<br/>S3-compatible · private + public buckets")]
    end

    subgraph ext["External services"]
        ai["AI providers<br/>OpenAI · Anthropic · Gemini"]
        tgapi["Telegram Bot API"]
        stripe["Stripe billing"]
    end

    web -->|"HTTPS"| https
    ios -->|"encrypted health sync API"| https
    tguser --- tgapi
    tgapi -->|"webhook updates"| https
    https --> app
    app -->|"enqueue AI jobs"| q
    q --> workers
    app --> db
    workers --> db
    app --> cache
    workers --> cache
    workers -->|"broadcast AI stream"| ws
    ws -.->|"live tokens + tool events"| web
    app & workers --> s3
    app & workers -->|"model + tool calls"| ai
    app -->|"bot replies"| tgapi
    app --> stripe

How a chat message moves through the system:

  1. The client sends the message over HTTPS to the web tier, which authenticates the request, persists the conversation, and enqueues an AI job so long-running agent work never blocks a web request.
  2. A queue worker runs the agent loop — health context assembly, tool calls, model requests — and broadcasts tokens, reasoning, and tool events over WebSockets as they stream.
  3. The browser renders the stream live, and the cache keeps resumable stream state so a dropped connection replays instead of losing the response. Telegram users reach the same agent through bot API replies.
  4. PostgreSQL remains the single source of truth for conversations and health entries, while object storage holds uploaded food photos and benchmark datasets.

Design principles:

  • The web tier is stateless, so replicas scale horizontally (or to zero) without session affinity; all state lives in PostgreSQL, the cache, and object storage.
  • AI work is async-first: every model call runs on the queue, which isolates provider latency and lets workers scale independently of web traffic.
  • Realtime is push-based over WebSockets rather than polling, with an event-replay fallback when a socket cannot connect.
  • Infrastructure roles are swappable bindings: the queue backend, cache, WebSocket server, and object store are configured through Laravel drivers, so deployments differ by configuration, not code.

Quick Setup

git clone https://github.com/acara-app/plate.git
cd plate
composer setup

composer setup installs PHP and JavaScript dependencies, creates .env when missing, generates the app key, runs migrations, installs Bun packages, and builds frontend assets.

Configure only the services you plan to use:

APP_URL=http://localhost

# Choose the AI providers you want to enable.
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
GEMINI_API_KEY=

# Optional Telegram integration.
TELEGRAM_BOT_TOKEN=
TELEGRAM_WEBHOOK_DOMAIN="${APP_URL}"

# Optional App Store override for local/private deployments.
HEALTH_SYNC_APP_STORE_URL=https://apps.apple.com/us/app/acara-health-sync/id6761504525

Start Redis with Docker for the real-time AI streaming features (no Herd Pro needed). PostgreSQL/pgvector run independently, so they are not in docker-compose.yml:

docker compose up -d

Run the development processes (app server, queue worker, Vite, logs, and the Reverb WebSocket server):

composer run dev

Run the project checks:

composer test

Useful targeted commands:

composer test:unit
composer test:types
composer test:lint
composer lint

Community and Premium Builds

The public repository should stay community-safe:

  • Do not require acara-app/plate-core for main.
  • Do not import private package namespaces in public app code.
  • Keep public extension contracts and null implementations in this repository so the app can boot without premium modules.
  • Use the private Plate Core package only in Acara private or Cloud deployments.

If you are working on the public app, install from main and treat premium behavior as optional integration surface, not a required dependency.

Health Sync Architecture

Acara Health Sync connects iPhone health data to Plate because web apps cannot read HealthKit directly.

The sync flow is:

  1. The user installs Acara Health Sync from the App Store.
  2. The iOS app authenticates with the user's Plate account.
  3. The app provisions secure sync credentials and stores them in the iOS Keychain.
  4. Health data is encrypted on device and sent to Plate through the sync API.
  5. Plate stores the readings and makes them available to dashboards and AI agent context.

The public Plate app owns the API and storage side of this flow. The iOS source repository is not public yet.

Telegram Chat Interface

Telegram support lets users interact with Acara from Telegram instead of opening the web app or PWA. Users can ask Acara the same health, nutrition, meal-planning, and tracking questions they ask in the web app, and the bot can parse natural language entries for glucose, food, carbs, insulin, medication, vitals, and exercise through the Telegram health logging tool, then confirm before saving.

For local webhook testing:

brew install ngrok
ngrok http https://plate.test
php artisan telegraph:new-bot
php artisan telegraph:set-webhook

Data and Health Context

Acara Plate uses structured profile data, health entries, conversation context, and user-approved device data to generate more relevant AI guidance, alongside application-specific glycemic context.

Food Photo Analysis

Plate reads food photos with a multimodal AI pipeline: the model identifies each item, estimates its portion, and estimates its calories and macros. You can review and edit every item before saving it. Results are useful estimates for awareness, not clinically validated measurements — never use them for insulin or medication dosing.

We measure how accurate the analysis really is with a first-party validation benchmark. See the data collection protocol.

Contributing

Contributions are welcome for the public Plate app. Start with the contributing guide, follow the code of conduct, and run the relevant tests before opening a pull request.

Good first areas to inspect:

  • AI agent tools and prompts
  • health data context and safety behavior
  • glucose and health dashboards
  • Apple Health sync API handling
  • Telegram chat interface
  • meal plan generation
  • image-based food analysis
  • translations and accessibility
  • public nutrition tools

License

Acara Plate is released under the O'Saasy License. It supports source review, modification, and self-hosting, with restrictions on offering the software as a competing hosted SaaS product.

Medical Disclaimer

Acara Plate provides AI-generated nutrition, wellness, and tracking support for informational and educational purposes only.

It is not professional medical advice, diagnosis, treatment, or a substitute for care from a licensed clinician. People with diabetes, prediabetes, pregnancy, eating disorders, kidney disease, cardiovascular conditions, medication changes, or other health concerns should consult a qualified healthcare professional before relying on any diet, medication, or glucose-management recommendation.

AI systems can make mistakes. Verify allergens, ingredients, medication details, glucose units, serving sizes, and nutrition values independently. In an emergency, contact local emergency services immediately.