๐ง Active Inference Institute โ Courses
Course catalogue ยท Module reference
Measured test facts
Minimize surprise. Maximize evidence.
This repository hosts Active Inference curricula spanning Kindergarten to PhD, a YouTube transcript archive and a GNN syntax track, powered by a modular Python publishing engine. See the catalogue and source-bound measurements rather than copied counts.
Maintained by Dr. Daniel Ari Friedman (@docxology) at the Active Inference Institute.
๐ Contents
๐ Curriculum Portfolio
All courses share a unified 8-topic spine grounded in the Free Energy Principle, revisited at increasing levels of mathematical and conceptual formalism.
| Topic Spine |
Systems โ Agents โ Perception โ Cognition โ Action โ Learning โ Communication โ Planning |
๐ Level-Adapted Tracks
| Course |
Audience |
Focus |
| Elementary |
Grades K-5 |
Stories, play, and drawing |
| Family |
Verified Families |
Parent-child co-learning activities |
| Middle School |
Grades 6-8 |
Scratch programming & logic |
| High School |
Grades 9-12 |
Python basics & guided labs |
| College 101 |
Undergraduates |
Full notation & standard simulations |
| Advanced 401 |
Graduate/PhD |
Research seminars & formal proofs |
๐ฌ Disciplinary Core (active_inference/)
| Track |
Focus |
Lab Style |
| Philosophy |
Argumentation & Ontology |
Thought Experiments |
| Cognitive Science |
Neuroscience & Psychology |
Case Studies |
| Mathematics |
Formal Derivations |
Proofs & Exercises |
| Computer Science |
Implementation & Algorithms |
Python active_inference Lib |
๐ Domain Applications
โก Quick Start
1. Prerequisites
- Python 3.11+
- uv (recommended)
2. Setup
git clone https://github.com/ActiveInferenceInstitute/courses.git
cd courses/software
uv sync
3. Render a Course
# Render the Philosophy track to Text and Markdown
uv run python scripts/generate_all_outputs.py --course ai-philosophy --formats txt,md
# Preview the full publishing pipeline
uv run python ../publish.py --dry-run
See the Quick Start Guide for full details.
๐ฆ Status
What works today, and what needs extra setup:
| Capability |
Status |
txt, md, html, docx rendering |
Works offline after cd software && uv sync. |
pdf rendering |
Needs system libraries (Cairo/Pango/GDK-PixBuf); see TROUBLESHOOTING.md. |
mp3 narration |
Needs network access for gTTS; disabled by default in publish.toml. |
| YouTube transcript archive |
Works from checked-in transcripts; live transcription needs yt-dlp/Whisper. |
| Translation and LLM enrichment |
Needs a running Ollama server. |
Canvas upload (canvas_integration) |
Needs Canvas API URL and token; not exercised by CI. |
| Offline test gate |
Runs in CI on Python 3.11 and 3.12. |
๐ Architecture & Software
This repository is powered by a custom Python rendering engine (software/) that orchestrates the transformation of Markdown source files into 6 output formats.
graph TD
Source["๐ Source Content<br/>(Markdown)"] --> Engine["โ๏ธ Publishing Engine<br/>(Python 3.11+)"]
Engine --> PDF["๐ PDF<br/>(WeasyPrint)"]
Engine --> HTML["๐ HTML<br/>(Interactive)"]
Engine --> DOCX["๐ DOCX<br/>(Word)"]
Engine --> MP3["๐ง MP3<br/>(gTTS)"]
Engine --> TXT["TXT"]
Engine --> MD["MD"]
style Source fill:#fef3c7,stroke:#d97706
style Engine fill:#dbeafe,stroke:#2563eb
style PDF fill:#d1fae5,stroke:#059669
๐ค Agent-Friendly Codebase
We maintain AGENTS.md files at every directory level to provide context-aware guidelines for AI agents.
โ
Gate
One entry point, one arm per check. CI calls these same arms, so a green local
run and a green CI step cannot diverge (docs/RUN_GUIDE.md):
./run.sh lint # ruff check + ruff format --check
./run.sh types # mypy src/
./run.sh no-mocks # no-mocks policy scan
./run.sh test # offline test gate (default)
./run.sh test-full # every test, including internet/API/whisper
./run.sh docs # relative links + generated facts current
./run.sh gnn # pinned GNN conformance
./run.sh all # the offline arms in order
Wall time per arm is measured, not estimated:
docs/_generated/ARM_TIMINGS.md.
๐ Documentation
Repository-level documentation โ the gate, CI, and the documentation map โ lives
in docs/. Engine documentation lives in
software/docs/, which indexes every technical guide:
๐ฌ YouTube Archive
We maintain a structured archive of ~821 video transcripts (38 playlists) from the Active Inference Institute's YouTube channel, organized into playlists including:
- Livestreams & Paper Discussions (171 videos)
- GuestStreams (142 videos)
- Textbook Cohorts (Parr et al. 2022)
- Applied Active Inference Symposia
See YouTube Documentation for the transcription and translation pipelines.
๐ค Contributing
We welcome contributions to both the curriculum and the software engine!
- Read CONTRIBUTING.md.
- Ensure you have
uv installed.
- Follow the Content Authoring guidelines.
Testing
The offline gate excludes internet, external-API and Whisper-required tests.
Use generated facts with their source/runtime
provenance; old counts and coverage do not qualify new source. The
run guide separates focused acceptance, the actual GNN
source-mode gate, installed-origin limits and the still-unproved full-network arm.
./run.sh test # the offline gate (default arm)
./run.sh test-full # every test, including the network-marked ones
๐ License
This work is licensed under CC BY 4.0.
ยฉ 2026 Active Inference Institute.
Generated by the Active Inference Institute Publishing Pipeline. Last updated: 2026-11-12.
InstituteOS sidecar
This repository carries a committed .aii/ sidecar with curated
InstituteOS metadata (aii-sidecar/v1): category, maturity, ecosystem relations,
key artifacts, provenance, and portable tasks. It is read by
InstituteOS tooling and
does not affect the project's build or runtime.