Tools
Brody AI ships 76 tools that the agent calls autonomously. Read-only tools run in parallel; writes stay sequential so the scene stays consistent.
Approval service
Destructive operations (deletes, rewrites, scene mutation) are gated behind explicit approval. Auto-approve mode keeps fully-automated runs moving while destructive tools stay confirmable — and rate limiting only kicks in below a healthy credit balance.
Scene building
Level design & enemy AI
Gameplay proof
Orchestration
Code & files
Package manager
Content & assets
Scene snapshots & rollback
Query & context
Memory
UI toolkit
Representative tools
build_game_auto
Executes a curated game preset end-to-end — dungeon_crawler, space_shooter, runner, fps_arena, rts_proto — or a genre keyword (platformer, fps, dungeon): spawns the template, builds the level layout, applies enemy AI plans, then runs verify_gameplay_loop as proof. One prompt, playable scene.
verify_gameplay_loop
Enters play mode and verifies the core loop with simulated input, reporting PASS/FAIL with evidence. Combined with auto_fix, this is the verify/fix loop behind every "verified" result.
spawn_subagents
The main agent calls this when a request is complex. It builds a 10-task plan across ten specialized roles — Planner, Coder, SceneBuilder, Verifier, Researcher, PhysicsArchitect, UIArchitect, AudioArchitect, VFXArchitect, NavMeshAIArchitect, and ErrorHandler — and dispatches them. It reports Spawned and dispatched 10 sub-agents. All verified. only when every task passed.
Scene snapshots
capture_scene_snapshot serializes the scene hierarchy and component state in memory; restore_scene_snapshot rolls back to it. Multi-step agent runs use this for clean rollback when a step fails.
Memory
memory_query recalls prior project state before planning;memory_sync persists meaningful changes so future sessions feel continuous. Backed by the memory service APIs — see the memory docs.
Project index & @-mentions
index_project builds a lightweight knowledge index of your C# scripts, prefabs, and scenes; mention_resolve turns @-mentions like @PlayerController into file context the chat can inject — with an @-mention symbol picker and tab-complete suggestions in the input box. risk_score_file scores any file 0–100 for how risky a change would be, and unity_docs_lookup answers Unity Scripting API questions from a local docs cache without hallucinating.
Playbooks
record_playbook starts/stops recording the current sequence of tool executions into Library/BrodyAI/playbooks/;replay_playbook replays a saved playbook step by step in the background, re-running the same tools with the same arguments. Reusable workflows — no re-prompting.
profile_play_mode
Enters play mode, samples native and managed memory every 0.5s, then exits — reporting peak allocated/reserved MB, GC heap, object count, and the top-5 largest memory growth deltas. Catch memory leaks before shipping.