Models · NeuralSpark 8.1
Your provider. Your key. Their invoice.
NeuralSpark does not sell AI credit. Bring a provider you already pay for, your Muse Code or ChatGPT plan, or a model running on this PC. Provider figures change, so check the provider page before you rely on one.
Plans you already have
No new bill for the plan in your pocket.
Pick a plan tile. NeuralSpark checks the app on this PC, names its version and account, and connects without sending a message to connect. What you use counts against that plan.
Muse Code plan
Through Muse Code, signed in on this PC. No API key. NeuralSpark never reads or holds the sign-in. It must be installed and signed in first.
ChatGPT plan, through Codex
Codex holds the address and format. Pick a model and connect. What the plan offers is set by that plan.
Model on this PC
Ollama, LM Studio or llama.cpp, with no key needed. Speed and quality depend on your PC and the model you run.

Real capture to comeDefault and backup
Hits a limit. Carries on. Comes back.
Set a default so new chats start right. Set a backup and, when a model reaches its limit or its provider is down, the chat moves by itself, tells you once, and goes back by itself. With no second model ready, it says so instead of failing quietly.
Backup model · Demo
One line appears above the message box: this conversation carries on with the backup until about a stated time, then goes back.
The same conversation, on the backup. Use counts against that provider or plan.
About an hour after a limit, about 15 minutes after an outage, or at the provider time. Go back now is there if you want it sooner.
Another provider, if it speaks the format.
OpenAI-compatible Chat Completions, OpenAI Responses, or Anthropic Messages. A wrong format is refused in plain words naming the format the model needs.
NeuralVault comes with it