If you want to learn active inference in 2026, you have more than one door. This is our map of the doors we have actually opened, written for someone who wants to touch the math and drive the model, not just read about it.
UNI is a working hypothesis on an attainable path toward General Natural Intelligence: a natural, active-inference approach whose evidence is growing, evidence-classed, and tested in the open. Do not take the claim on faith. Test the build, inspect the gates, and help us find where it fails.
The math and the textbook
Start with Parr, Pezzulo, and Friston (2022), Active Inference: The Free Energy Principle in Mind, Brain, and Behavior, MIT Press Class E. It is the current canonical treatment: variational free energy as an upper bound on surprise, POMDP formulation of perception and action, expected free energy as the policy criterion, and precision as the modulation knob on prior, likelihood, and policy. Every other resource on this page assumes the reader can meet this book at least halfway.
Namjoshi (2026), the Stratified Palimpsest benchmark, is the newer reference we cite for precision-under-disturbance testing Class E. Our Cell Lab is aligned with its evaluation posture: pre-register the falsification criteria, run the disturbance families, publish the losses next to the wins.
Five doors, one field
1. The Themesis resource map
Where to Start with Active Inference, A Resource Map for 2026 (Themesis) is the map that lists SolutionWright Universal among five pathways for people entering the field Class E. We use it as the anchor for our family tie-ins: if someone finds us through that map, this post is where the arrow lands.
2. Hands-on Python
Building Active Inference in Python (Themesis) is a complementary hands-on Python course, a different stack than our Elixir workbench Class E. If you want to hold the math with pip install and a notebook, this is a good door.
3. Deep-dive video
The Themesis YouTube channel has deep-dive videos on active inference we recommend to learners who prefer to watch a topic unfold before reading it Class E. Pair it with the textbook: one chapter, one video, one lab session.
4. Steerable browser labs
Our own Precision Lab, Echo Lab, Loop Lab, Heart Lab, and Cell Lab are five instruments that run in the browser with no backend Class C. You move the sensory-precision, transition-precision, and policy-temperature dials, and watch Bayesian inference change shape in front of you. The Cell Lab is a pre-registered falsification benchmark: a UNI active-inference controller against random, rule-based, and neural baselines on a hidden 216-state service cell under disturbance.
5. The MCP surface for language models
Anyone building an LLM agent can call the labs directly over Model Context Protocol at universalnaturalintelligence.com/api/mcp Class C. Sixteen tools, no auth, just the URL. It is the door for machines.
How to walk the map
We suggest one path for newcomers: read the textbook front matter, open the Precision Lab in a tab, and read one chapter with the lab already breathing beside you. When the Markov blanket and the KL divergence stop being words on a page, come back here and add the Cell Lab. When the Cell Lab surprises you, and it will, the disturbance-family losses are the interesting part, not the wins.
We are not the only door, and we do not want to be. The field is stronger with more entry points, and the honest move is to name the ones we have used.