Environment variables for _AugmentedIntelligence

Recommended defaults for a dev workstation. Empty string means “unset / feature off until configured.”

Also see: RemoteCUDA.md · BuildConfigurations.md · PluginBridges.md

Quick start (dev defaults)

REM --- Core (usually leave unset; Settings/MySQL file wins) ---
REM set AI_MYSQL_HOST=127.0.0.1
REM set AI_MYSQL_USER=root
REM set AI_MYSQL_PASSWORD=
REM set AI_MYSQL_DATABASE=cognitive_loop

REM --- Remote CUDA worker (optional GPU box) ---
set AI_REMOTE_CUDA_ENABLED=1
set AI_REMOTE_CUDA_HOST=127.0.0.1
set AI_REMOTE_CUDA_PORT=7720
set AI_REMOTE_CUDA_TOKEN=dev-token
set AI_REMOTE_CUDA_TIMEOUT_MS=15000

REM On the GPU machine:
REM   set AI_REMOTE_CUDA_TOKEN=dev-token
REM   set AI_REMOTE_CUDA_PORT=7720
REM   python tools\remote_cuda_worker.py

Remote CUDA / HardwareAcceleration

In-app Settings globals (not env, but related):
cuda_accelleration_remote_enabled, cuda_accelleration_remote_hostname,
tensor_accelleration_remote_*, hardware_remote_rpc_port (default 7720),
hardware_remote_rpc_timeout_ms (15000), hardware_remote_rpc_token.

Worker: python tools/remote_cuda_worker.py
Client commands: cuda remote status|ping|hello|bench|set|job|tensor|checkpoint|…

Database / MySQL

Application settings load/save via MySQL ai_settings by default. See MySQLConfiguration.md.

If you use env-based secrets tooling, prefer OS Credential Manager / GPO (_Settings::ApplyAdPasswords) over plain env in production.

LLM / remote inference (non-CUDA)

Bridges / games

No required env for bridges; dry-run on by default in BridgeManager. Creo/Blender/C4D work offline with *_API=mock (default). Weather uses Open-Meteo + NWS + USGS — see WeatherHazards.md.

Plugins

Sandbox paths: plugins/data/ (HostApi FS).

CUDA / GPU local

Build-time: USE_CUDA / AI_CPU_ONLY in VS configurations (not runtime env).

Networking / portal

Testing / CI

Recommended profiles

A. Laptop CPU-only

set AI_REMOTE_CUDA_ENABLED=0
REM do not set AI_REMOTE_CUDA_HOST

B. Local GPU + local worker (same machine)

set AI_REMOTE_CUDA_ENABLED=1
set AI_REMOTE_CUDA_HOST=127.0.0.1
set AI_REMOTE_CUDA_PORT=7720
set AI_REMOTE_CUDA_TOKEN=dev-token

C. Workstation + remote GPU server

set AI_REMOTE_CUDA_ENABLED=1
set AI_REMOTE_CUDA_HOSTS=http://10.0.0.50:7720,http://10.0.0.51:7720
set AI_REMOTE_CUDA_TOKEN=prod-long-random-token
set AI_REMOTE_CUDA_TIMEOUT_MS=30000

D. Match HardwareAcceleration remote Tensor path

set AI_REMOTE_CUDA_HOST=10.0.0.50
set AI_REMOTE_CUDA_PORT=7720
set AI_REMOTE_CUDA_TOKEN=same-as-hardware_remote_rpc_token

Also enable remote CUDA/tensor hostnames in Settings if you use the older HA code paths.

Security notes

1. Never commit real tokens. Use dev-token only on localhost.
2. Bind workers to LAN or localhost; use firewall + strong token in production.
3. Kill-switch / safe mode cancels remote CUDA jobs.
4. Remote matmul always falls back to local GEMM if the worker is down.

Verification

cuda remote refresh
cuda remote status
cuda remote ping
cuda remote bench 128
nn device list
nn device set global train remote

Worker logs should show /v1/ping and /infer or /v1/matmul hits.