2026.10.13

AGI Pages

agi-pages is the umbrella/provider package for AGILAB public analysis views. It exposes a small discovery API so the AGILAB ANALYSIS page and exported notebooks can resolve installed view bundles without embedding the page source in the root agilab wheel.

The native UI described here is the current source implementation. Previously published PyPI packages may still use the former provider; the commands below install their published release, not this source migration. From an AGILAB source checkout, use uv sync --extra ui --extra notebook to install the native interface and notebook widget. Local validation does not publish those changes.

Quick Install

pip install agi-pages

Most users should install it through the AGILAB UI profile:

pip install "agilab[ui]"

Runtime Contract

The package exposes a small provider API:

import agi_pages

print(agi_pages.bundles_root())

agi-env uses this provider only when agi-pages is installed. A base agilab install remains CLI/core-only and does not require this package.

The default umbrella covers the shared AGILAB UI runtime. Specialized views with narrower runtime constraints are published independently as agi-page-* wheel/sdist payload packages. Install the page package you need alongside agi-pages when a notebook or app depends on that view.

Portable Chart Specs

agi-pages also exposes a lightweight chart contract for page bundles that need the same dataframe-driven visualization to work in native Python UI, exported notebooks, and static proof artifacts.

import agi_pages

spec = agi_pages.build_chart_spec(
    [{"step": "train", "accuracy": 0.84}, {"step": "test", "accuracy": 0.82}],
    chart_type="line",
    title="Model metrics",
    x="step",
    y="accuracy",
)

agi_pages.render_streamlit(spec)  # compatibility name; renders with agi_web.python_ui

The spec stores an ECharts-compatible option, normalized table records, and a small evidence block with deterministic hashes for the data, chart option, and combined chart contract. Page bundles can write spec.as_dict() into a proof artifact, render agi_pages.render_notebook(spec) from an exported notebook, or embed agi_pages.chart_spec_to_static_html(spec) in a standalone HTML report.