TreeQuest

Why it's ranked here

TreeQuest is a focused research-engineering library with unusually clear separation between search policy and model execution. Multiple Monte Carlo variants, order-independent result reporting, checkpoint support, and inspectable visualizations make it credible for controlled inference experiments. The heavier mixed-model path and scoring requirements keep it from being a turnkey answer system.

What's good

States can be any Python object, and actions can represent different models, prompts, or strategies. The ask-and-tell workflow separates candidate generation from tree management, supports batches, accepts results in any order, and ignores duplicate reports. Search state lives outside algorithm objects, which makes persistence and orchestration easier.

Tradeoffs

Users must implement generation, evaluation, and normalized scoring, so usefulness depends heavily on an external judge. The mixed-model algorithm needs substantial optional scientific dependencies and can run slowly. Large batches may produce overly wide trees, while HTML visualization requires careful handling when state content is untrusted.

How to use it well

Use TreeQuest when Python researchers or inference engineers need explicit control over branching, budgets, models, prompts, and scoring. Start with small batches, persist the returned search state, and inspect tree outputs while tuning policies. It does not supply an LLM provider, prompts, or a trustworthy evaluator.

Observed

License

Apache 2.0

Primary language

Python 3.11 or newer

Packaging

Hatchling package installable with pip or uv

Interface

Python library with direct and batched ask-and-tell search workflows

Optional dependencies

Separate extras for mixed-model AB-MCTS and visualization

Algorithms

AB-MCTS-A, AB-MCTS-M, Standard MCTS, and Tree of Thoughts breadth-first search are exposed

Visualization

Supports Graphviz images, JSON, YAML, Mermaid, Markdown, and interactive HTML output