Skip to content

See docs/DISCLAIMER_SNIPPET.md

Project Documentation

α‑AGI Insight Demo

Deployment Quickstart

  • Browser Quickstart – run ./scripts/deploy_insight_full.sh for a verified one-command deployment
  • One‑Command Deployment – execute ./scripts/insight_sprint.sh to build, verify and publish the GitHub Pages site automatically.
  • Local Gallery Build – run ./scripts/build_gallery_site.sh to compile the full demo gallery under site/ for offline review.
  • Build & Open Gallery – run ./scripts/build_open_gallery.sh to regenerate the docs and open the gallery automatically.
  • Preview Gallery Locally – run ./scripts/preview_gallery.sh to build the full gallery and serve it on http://localhost:8000/.
  • Open Gallery (Python) – run ./scripts/open_gallery.py for a cross-platform way to launch the published gallery, falling back to the local build when offline.
  • Open Gallery (Shell) – run ./scripts/open_gallery.sh to open the gallery in your browser. It automatically builds a fresh local copy when the remote site isn't available.
  • Open Individual Demo – run ./scripts/open_demo.sh <demo_dir> to open a single page from the gallery.
  • Open Subdirectory Gallery – run ./scripts/open_subdir_gallery.py to launch the mirror under alpha_factory_v1/demos/.
  • Open Subdirectory Demo – run ./scripts/open_subdir_demo.py <demo_dir> to open a single page from that mirror.
  • Offline Tests – build a wheelhouse with scripts/build_offline_wheels.sh then run python check_env.py --auto-install --wheelhouse <dir> and pytest without network access.

Building the React Dashboard

The React dashboard sources live under alpha_factory_v1/demos/alpha_agi_insight_v1/src/interface/web_client. Build the static assets before serving the API:

npm ci --prefix alpha_factory_v1/demos/alpha_agi_insight_v1/src/interface/web_client
npm --prefix alpha_factory_v1/demos/alpha_agi_insight_v1/src/interface/web_client run build

The compiled files appear in alpha_factory_v1/demos/alpha_agi_insight_v1/src/interface/web_client/dist and are automatically served when running uvicorn alpha_factory_v1.demos.alpha_agi_insight_v1.src.interface.api_server:app with RUN_MODE=web.

Ablation Runner

Use alpha_factory_v1/core/tools/ablation_runner.py to measure how disabling individual innovations affects benchmark performance. The script applies each patch from benchmarks/patch_library/, runs the benchmarks with and without each feature and generates docs/ablation_heatmap.svg.

python -m alpha_factory_v1.core.tools.ablation_runner

The resulting heatmap visualises the pass rate drop when a component is disabled.

Manual Workflows

The repository defines several optional GitHub Actions that are disabled by default. They only run when the repository owner starts them from the GitHub UI. These workflows perform heavyweight benchmarking and stress testing.

To launch a job:

  1. Open the Actions tab on GitHub.
  2. Choose either 📈 Replay Bench, 🌩 Load Test or 📊 Transfer Matrix.
  3. Click Run workflow and confirm.

Each workflow checks that the person triggering it matches github.repository_owner, so it executes only when the owner initiates the run.

Macro-Sentinel Demo

A self-healing macro risk radar powered by multi-agent α‑AGI. The stack ingests macro telemetry, runs Monte-Carlo simulations and exposes a Gradio dashboard. See the alpha_factory_v1/demos/macro_sentinel/README.md for full instructions.

α‑AGI Insight v1 Demo

docs/alpha_agi_insight_v1 provides a self-contained HTML demo that visualises capability forecasts with Plotly. The GitHub Actions workflow copies this directory into the generated site/ folder, serves it on GitHub Pages and deploys the page automatically. Visit the published demo to preview it.

The old static_insight directory has been removed in favour of this official static demo.

To update the charts, edit forecast.json and population.json and rebuild the site:

./scripts/edge_human_knowledge_pages_sprint.sh

This helper fetches all assets, compiles the browser bundle and runs mkdocs build. Open site/alpha_agi_insight_v1/index.html in your browser to verify the changes before committing. Alternatively run ./scripts/preview_insight_docs.sh to build and serve the demo locally on http://localhost:8000/.