metatotai 0.1.0

Active-inference-driven meta tree-of-thought planning engine. Public protocol surface (LLMProvider, TraceSink, WorldModelProvider, MemoryProvider, ThoughtseedPromoter, DecisionLearner). Engine implementation arrives in 0.2.0. Distinct from kyegomez/Meta-Tree-Of-Thoughts.

metatotai

Status: name reserved. Real implementation arrives in 0.1.0.

metatotai will be an active-inference-driven meta tree-of-thought planning engine — a deterministic, replay-safe decision layer that consumes elume's cognitive substrate and linoss-dynamics physics primitives.

What this will be

• Active-inference engine — variational free energy + expected free energy computation, belief updates, policy scoring.
• Meta tree-of-thought planner — search and expansion over reasoning trajectories, scored by free-energy minimization rather than LLM-graded floats.
• Theory of mind — PartnerModel for inferring other agents' beliefs from observed behavior, built on elume.MentalModel.
• Provider-injected — LLMProvider, MemoryProvider, WorldModelProvider, TraceSink protocols. Bring your own backend.

What this is not

• Not a fork of kyegomez/Meta-Tree-Of-Thoughts. That is a LangChain-based prompt-rewriting meta-agent over LLM scoring; this is a different system. Zero shared code. Distinct name to avoid confusion.
• Not the original Tree of Thoughts (Yao et al. 2023). Different mechanism.

Layering

linoss-dynamics  ← physics primitive (NumPy)
       ↑
   elume         ← cognitive substrate (mental models, basins, evolution)
       ↑
   metatotai     ← active-inference + meta-ToT planning

Install (when 0.1.0 ships)

pip install metatotai

Citations

metatotai's active-inference implementation adopts the conceptual framing from Karl Friston's free-energy principle. The implementation is original Python code; conceptual framing only is shared with upstream sources.

Foundational active inference:

• Friston, K. J. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138. https://doi.org/10.1038/nrn2787
• Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., & Pezzulo, G. (2017). Active Inference: A Process Theory. Neural Computation, 29(1), 1–49. https://doi.org/10.1162/NECO_a_00912
• Parr, T., Pezzulo, G., & Friston, K. J. (2022). Active Inference: The Free Energy Principle in Mind, Brain, and Behavior. MIT Press. https://mitpress.mit.edu/9780262045353/

Reference implementations consulted (no shared code):

• pymdp — Heins, C., Millidge, B., Demekas, D., Klein, B., Friston, K., Couzin, I. D., & Tschantz, A. (2022). pymdp: A Python library for active inference in discrete state spaces. Journal of Open Source Software. https://github.com/infer-actively/pymdp (MIT)
• ActiveInference.jl — Nehrer, S. M. et al., Julia package for active inference and POMDP modeling. https://github.com/samuelnehrer02/ActiveInference.jl (MIT)

Tree-of-thought lineage:

• Yao, S., Yu, D., Zhao, J., Shafran, I., Griffiths, T. L., Cao, Y., & Narasimhan, K. (2023). Tree of Thoughts: Deliberate Problem Solving with Large Language Models. NeurIPS 2023. https://arxiv.org/abs/2305.10601

Substrate dependencies (used at runtime):

• elume — cognitive substrate (mental models, basins, evolution)
• linoss-dynamics — LinOSS physics primitive (Rusch & Rus, ICLR 2025)

Distinct from (zero shared code):

• kyegomez/Meta-Tree-Of-Thoughts — LangChain-based prompt-rewriting meta-agent over LLM scoring. Different mechanism, different stack. Distinct name to avoid academic and licensing confusion.

License

MIT.