Minimal App Project

Overview

• Minimal starter template you can copy to bootstrap a new AGILab application.

• Demonstrates the project layout expected by the platform (manager package, worker package, app_args definitions, Analysis configuration) with minimal business logic so you can focus on custom code.

• Ships with a small custom Streamlit argument form and an empty pre_prompt.json prompt seed so copied projects have explicit places for UI customisation and WORKFLOW prompts.

Scientific placeholders

The template is intentionally lightweight, but many AGILab workflows ultimately fit the pattern of learning or calibrating a function \(f_\theta\):

\[\theta^* = \arg\min_{\theta} \; \frac{1}{N} \sum_{i=1}^{N} \ell\left(f_{\theta}(x_i), y_i\right)\]

where \(\ell\) is a task-dependent loss (regression, classification, imitation learning, etc.). You can use the Minimal App skeleton to prototype the data loading, feature extraction, and artifact export around this core loop.

Manager (minimal_app.minimal_app)

• Lightweight BaseWorker subclass that prepares app-owned shared input and output directories through env.resolve_share_path.

• Accepts validated MinimalAppArgs values, supports reset_target cleanup of the output directory, and keeps from_toml / to_toml helpers for the same settings round-trip used by richer apps.

• Leaves work distribution as a template hook: build_distribution currently returns empty placeholder structures until a copied project adds real stages.

Args (minimal_app.app_args)

• Simple Pydantic model with defaults for data_in, data_out, files, nfile, nskip, nread, and reset_target.

• src/app_settings.toml intentionally seeds an empty [args] table; the runtime loads model defaults, then persists concrete values only after the custom src/app_args_form.py edits them.

• The model still migrates legacy data_uri input into data_in so older snippets can be adapted without changing the scaffold shape first.

Worker (minimal_app_worker.minimal_app_worker)

• Concrete MinimalAppWorker subclass of agi_node.polars_worker.PolarsWorker.

• It intentionally adds no domain logic; the class exists so installers and worker deployment can discover a real worker package while copied projects replace the inherited placeholder behavior with domain-specific processing.

Reducer contract status

minimal_app_project is template-only. It intentionally does not ship a reducer contract because the manager distribution hook and worker behavior are scaffold placeholders and no concrete merge output exists yet. This is an explicit reducer exemption for the starter template, not a gap in a promoted domain app.

When a cloned project starts producing durable worker summaries, add a reduction.py module, emit reduce_summary_worker_<id>.json artifacts, and export a *_REDUCE_CONTRACT symbol from the manager package. That keeps custom apps aligned with the shared agi_node reducer contract without treating the starter template as an unfinished built-in app.

Assets & Tests

• README.md and pyproject.toml describe the installable project package and its runtime dependencies.

• src/app_settings.toml carries cluster defaults plus the intentionally empty [args] seed.

• src/app_args_form.py is the current custom ORCHESTRATE form for path and small run-control arguments; tailor it when you need additional validation or widgets.

• src/pre_prompt.json is currently an empty prompt list.

• Repo-level tests exercise the installer, clone path, runtime bootstrap, and minimal app environment behavior rather than project-local test/_test_* files.

API Reference

Minimal manager implementation for the minimal_app sample project.

class minimal_app.minimal_app.MinimalApp(env, args=None, **kwargs)

Bases: BaseWorker

Lightweight orchestration surface for the minimal_app example.

__init__(env, args=None, **kwargs)

as_dict()

Return type:

dict[str, Any]

build_distribution(_workers=None)

Return type:

Tuple[List[List], List[List[Tuple[int, int]]], str, str, str]

classmethod from_toml(env, settings_path='app_settings.toml', section='args', **overrides)

Return type:

MinimalApp

static pool_init(vars)

Per-pool-child initializer hook; default is a no-op.

Return type:

None

stop()

Returns:

Return type:

None

to_toml(settings_path='app_settings.toml', section='args', create_missing=True)

Return type:

None

work_done(_)

Return type:

None

work_pool(_=None)

Return type:

None

worker_vars = {}

class minimal_app.minimal_app.MinimalAppApp(env, args=None, **kwargs)

Bases: MinimalApp

Alias retaining the historical suffix for compatibility.

Argument management helpers for the minimal_app sample project.

minimal_app.app_args.ArgsModel

alias of MinimalAppArgs

minimal_app.app_args.ArgsOverrides

alias of MinimalAppArgsTD

class minimal_app.app_args.MinimalAppArgs(**data)

Bases: BaseModel

Runtime parameters for the minimal_app application.

data_in

data_out

files

model_config = {'extra': 'forbid'}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].

nfile

nread

nskip

reset_target

class minimal_app.app_args.MinimalAppArgsTD

Bases: TypedDict

data_in

data_out

files

nfile

nread

nskip

reset_target

minimal_app.app_args.dump_args(args, settings_path, *, section='args', create_missing=True)

Return type:

None

minimal_app.app_args.ensure_defaults(args, **_)

Return type:

MinimalAppArgs

minimal_app.app_args.load_args(settings_path, *, section='args')

Return type:

MinimalAppArgs

minimal_app.app_args.merge_args(base, overrides=None)

Return type:

MinimalAppArgs

Default worker implementation for the minimal_app project.

This worker simply inherits the PolarsWorker so that projects with minimal requirements still provide a concrete worker class. Downstream installers rely on the class name MinimalAppWorker to determine the runtime bundle to ship.

class minimal_app_worker.minimal_app_worker.MinimalAppWorker

Bases: PolarsWorker

Polars worker used by the minimal_app sample application.