MathNN multi-level access to _AI features

MathNN is not a single toy net: it can touch many layers of _AugmentedIntelligence. The MathNN Access Hub (MathNNAccessHub) catalogs those paths, gates levels/features, and gives probe / recipe helpers so agents and operators can wire learning into the right subsystem.

Levels (low → high abstraction)

Caps on each feature: encode · train · act · observe · speak.

Commands

nn access status
nn access levels
nn access map
nn access map 3 4              # agents + world only
nn access list agents
nn access feature agent_minecraft
nn access probe simple_fuse
nn access recipe minecraft_dqn
nn access recipe brainstorm_stack
nn access probe brainstorm_stack
nn access agent minecraft
nn access enable level 5
nn access disable feature gaming_ethics_gate
nn access enable feature agent_minecraft

Also: nn status prints a one-line access summary.

Lua

print(mathnn_access_status())
print(mathnn_access_map(0, 5))
print(mathnn_access_feature("agent_minecraft"))
print(mathnn_access_probe("simple_fuse"))
print(mathnn_access_recipe("math_algebra"))
print(mathnn_access_enabled("minecraft_dqn"))

Registered from _MathNNFeatures::RegisterLuaGlobals / agent Lua runtime.

Example wiring recipes

Minecraft MathNN policy

nn access recipe agent_minecraft
rl net enable minecraft dueling-dqn
minecraft bridge start
minecraft dqn net on
minecraft dqn train 1000

Simple multimodal → neural gate

simple nn encode text hello world
simple nn fuse
simple nn predict
nn access probe simple_gate

Math tutoring domain

nn math domains
nn math train algebra 20
nn math tutor step algebra

Files

Access factory (actors)

Create scoped handles for actors/agents so each bot only gets the MathNN levels it needs.

nn access factory presets
nn access factory create minecraft mc_bot Steve
nn access factory create tutor math_tutor_1
nn access factory create observe watcher
nn access factory list
nn access factory describe mc_bot
nn access factory bind mc_bot Steve
nn access factory can mc_bot minecraft_dqn act
nn access factory can mc_bot llm_coach train
nn access factory destroy mc_bot

Lua

local ok, err = mathnn_factory_create("minecraft", "mc_bot", "Steve")
print(ok, err)
print(mathnn_factory_describe("mc_bot"))
local can, why = mathnn_factory_can("mc_bot", "minecraft_dqn", "act")
print(can, why)
print(mathnn_factory_list())

Authorization checks: global feature/level gate and handle allow-list and capability flags (encode|train|act|observe|speak).

Design notes

• Default: all levels enabled. Use nn access disable level|feature to sandbox.
• Hub is a map + gate + recipes, not a second training engine. Live training still goes through domain/agent commands.
• Factory handles are in-memory actor sandboxes; use them from agents/Lua before calling train/act.
• Probes call cheap Status() APIs when linked; otherwise they print the primary command hint.
• Extend the catalog in seedFeatures() inside MathNNAccessHub.cpp when new _AI subsystems should be MathNN-reachable.

Related

• MathNN.md
• MathNNFields.md
• MathNNLearningFunctions.md
• SimpleMathNNBridge.md
• MathNNCoachTrain.md
• ReinforcementLearning.md