marona 0.4.1

Use Marona to build AI apps for any AI interface, edge device, or intelligent agent. Build once and deploy online, offline, or in hybrid environments through a unified runtime.

marona

Official Python client for Marona-compatible AI runtimes and Hub integrations.

Use Marona to build AI apps for any AI interface, edge device, or intelligent agent. Build once and deploy online, offline, or in hybrid environments through a unified runtime.

Install

pip install marona

Complete Example

This example shows the normal flow for a developer application:

1. Create a Marona client.
2. Sync Hub metadata into the local cache.
3. Connect approved Hub apps by app ID.
4. Send a user message through the runtime.
5. Print the final assistant response.

The optional developer role controls app behavior.

import asyncio
import os

from marona import Marona


async def main() -> None:
    api_key = os.environ["MARONA_API_KEY"]
    identity_token = os.getenv("MARONA_IDENTITY_TOKEN")  # Optional

    async with Marona(api_key=api_key, mode="online") as marona:
        await marona.sync(
            interface="api",
            identity_token=identity_token,
        )

        tools = await marona.hub.connect(
            ["sda-books", "zimsec"],
            adapter="tools",
        )

        response = await marona.client(
            input=[
                {
                    "role": "developer",
                    "content": "Keep answers clear and concise.",
                },
                {
                    "role": "user",
                    "content": [
                        {
                            "type": "input_text",
                            "text": "What teams are playing in this image?",
                        },
                        {
                            "type": "input_image",
                            "image_url": "https://api.nga.gov/iiif/a2e6da57-3cd1-4235-b20e-95dcaefed6c8/full/!800,800/0/default.jpg",
                        },
                        {
                            "type": "input_file",
                            "filename": "document.pdf",
                            "file_data": "data:application/pdf;base64,...",
                            "detail": "high",
                        },
                    ],
                }
            ],
            interface="api",
            identity_token=identity_token,
            tools=tools,
        )

        print(response.text)


asyncio.run(main())

Build And Publish Skills

Skills define reusable, ordered workflows over Apps. Publishing validates step IDs, references, required Apps, exact capabilities, and the immutable version.

import os

from marona import Marona
from marona.skills import skill, step


@skill(
    name="create-group-fund",
    description="Create a group fund after explicit user approval.",
    governs=["group-fund.create_group"],
)
def create_group_fund():
    request = step(
        id="understand-request",
        type="reasoning",
        instruction="Extract the group name and currency.",
        inputs={"message": "{{ context.user_message }}"},
        outputs={"name": "string", "currency": "string"},
    )
    permission = step(
        id="confirm-create",
        type="approval",
        message=f"Create '{request.name}' in {request.currency}?",
        outputs={"approved": "boolean"},
    )
    return step(
        id="create-group",
        type="app",
        app="group-fund",
        capability="group-fund.create_group",
        instruction="Create the approved group.",
        condition=permission.approved,
        inputs={"name": request.name, "currency": request.currency},
        outputs={"group_id": "string", "name": "string"},
    )


marona = Marona(api_key=os.environ["MARONA_API_KEY"])
marona.skills.publish(create_group_fund, version="1.0.0")

marona.sync() synchronizes published Apps and Skills. Users still call marona.client(...); the model discovers a matching Skill and the runtime enforces its ordered steps and approvals. There is no separate Skill run API. governs declares raw capabilities that must only run through that workflow. Every workflow entry uses step(); its type selects reasoning, approval, or App execution. There are no separate type-specific step builders.

The model selects the best matching Skill from the synced catalog. Marona, not the model provider, executes its ordered step() values, persists approval gates, and invokes each exact MCP capability. Offline exposes only Skills whose required Apps have installed offline/hybrid targets; governed capabilities are never exposed as raw tools. Hybrid falls back to Edge when the local route cannot produce a valid result.

Run it:

export MARONA_API_KEY="mrn_live_..."
python app.py

Pair A User

Use pairing when an interface needs to become the same user across web, mobile, WhatsApp, wearables, or another client surface.

async with Marona(api_key=os.environ["MARONA_API_KEY"]) as marona:
    pairing = await marona.start_pairing(
        interface="web",
        device_name="Customer web chat",
    )

    print(pairing.display_code)
    print(pairing.whatsapp_url)

    status = await marona.pairing_status(pairing.pairing_id)
    print(status.status)

After the user confirms pairing, store the returned identity token and pass it to marona.client(...).

Online, Offline, And Hybrid Modes

Choose one runtime mode for the client:

• online: use online runtime and online app routes.
• offline: use only local cache, local/private models, and installed offline-capable app targets.
• hybrid: try local/private execution first, then use online runtime when allowed.

Apps also declare one availability mode:

• online: online only.
• offline: offline only.
• hybrid: online and offline capable.

Hybrid is a single mode. Do not declare online + offline + hybrid; declare hybrid.

Hybrid Or Offline With A Local/Private Model

Marona uses one provider-neutral runtime flow. Registering a model changes the inference provider; App/Skill discovery, MCP execution, context, permissions, and hybrid fallback remain unchanged.

Register a local or private model endpoint once, then use it as the default for hybrid or offline execution.

async with Marona(api_key=os.environ["MARONA_API_KEY"], mode="hybrid") as marona:
    await marona.sync(interface="api")

    marona.models.register(
        name="office-model",
        endpoint="http://localhost:9379",
    )
    marona.models.use("office-model")

    tools = await marona.hub.connect(["sda-books", "zimsec"], adapter="tools")

    response = await marona.client(
        input=[
            {
                "role": "user",
                "content": [
                    {
                        "type": "input_text",
                        "text": "Summarize the chapter about faith.",
                    }
                ],
            }
        ],
        interface="desktop",
        tools=tools,
    )

    print(response.text)

In offline mode, Marona never calls public cloud runtime. If a model or an offline-capable app target is missing, the client returns a clear offline error. Only installed offline/hybrid Apps and Skills whose required Apps are available are discoverable. Skill-managed capabilities are not exposed as raw App tools.

Synced conversation context and connected tool schemas are supplied to the configured model on every turn. The current user prompt remains the active task, and the model decides whether to answer directly or call an available tool.

Interface Names

interface identifies the client surface making the request. Standard values:

• api
• web
• mobile_app
• desktop
• whatsapp

Future devices can use custom lowercase slugs such as smart_glasses, vehicle_console, or kiosk.

Related Packages

• Python client: marona, import marona
• Dart client: marona, import package:marona/marona.dart
• TypeScript client: marona, import Marona from "marona"
• Developer SDK for building apps: marona-sdk, import marona_sdk