What it does

big-brain gives your Bevy entities a decision loop: Scorers read the world and return numeric scores, Thinkers pick the winning option, and Actions execute it. Everything is plain Rust, data-driven, and wired through Bevy’s ECS as ordinary components and systems. You write the domain logic; the library handles the plumbing.

The interesting bit

The clever part is how deeply it leans into Bevy’s architecture. Scorers run in parallel as ECS systems. Actions are state machines with explicit cancellation hooks. The Thinker itself is just another component you spawn. It feels less like bolting on an AI framework and more like extending your game’s entity system to think.

Key highlights

• Scorers, Actions, and Thinkers compose as regular Bevy components and systems
• Concurrent scorer evaluation via ECS parallelism
• State-machine actions with Requested → Executing → Success/Failure/Cancelled lifecycle
• Reflect trait support for debugging with inspectors (explicitly unstable, debug-only)
• Compatible with Bevy 0.16.0; MSRV tracks Bevy’s latest stable

Caveats

• Active development has moved to Codeberg; this GitHub repo is a read-only archive that may disappear
• Reflection support is deliberately not semver-stable; fields can change without warning

Verdict

Worth a look if you’re building a Bevy game and want AI that reasons about trade-offs rather than following rigid trees. Skip it if you’re not on Bevy, need guaranteed long-term GitHub hosting, or want a more mature ecosystem with broader community support.